From 6761719387c6effc2db0e297433cd7010164ffaa Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 13:48:37 +0900
Subject: [PATCH 01/22] [feat] Support faster-whisper with adapter

---
 .../recognizers/whisper_local/base.py         |  4 -
 .../whisper_local/faster_whisper.py           | 45 +++++++++++
 .../recognizers/whisper_local/whisper.py      | 18 ++++-
 .../whisper_local/test_faster_whisper.py      | 75 +++++++++++++++++++
 4 files changed, 137 insertions(+), 5 deletions(-)
 create mode 100644 speech_recognition/recognizers/whisper_local/faster_whisper.py
 create mode 100644 tests/recognizers/whisper_local/test_faster_whisper.py

diff --git a/speech_recognition/recognizers/whisper_local/base.py b/speech_recognition/recognizers/whisper_local/base.py
index 7bf458b6..12e04e65 100644
--- a/speech_recognition/recognizers/whisper_local/base.py
+++ b/speech_recognition/recognizers/whisper_local/base.py
@@ -32,10 +32,6 @@ def recognize(
         audio_array, sampling_rate = sf.read(wav_stream)
         audio_array = audio_array.astype(np.float32)
 
-        if "fp16" not in kwargs:
-            import torch
-
-            kwargs["fp16"] = torch.cuda.is_available()
         result = self.model.transcribe(audio_array, **kwargs)
 
         if show_dict:
diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
new file mode 100644
index 00000000..58d4b860
--- /dev/null
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+from typing import TYPE_CHECKING, TypedDict
+
+from speech_recognition.audio import AudioData
+from speech_recognition.recognizers.whisper_local.base import (
+    WhisperCompatibleRecognizer,
+)
+
+if TYPE_CHECKING:
+    from faster_whisper import WhisperModel
+    from faster_whisper.transcribe import Segment
+
+
+class TranscribeOutput(TypedDict):
+    text: str
+    segments: list[Segment]
+    language: str
+
+
+class TranscribableAdapter:
+    def __init__(self, model: WhisperModel) -> None:
+        self.model = model
+
+    def transcribe(self, audio_array, **kwargs) -> TranscribeOutput:
+        segments_generator, info = self.model.transcribe(audio_array)
+        segments = list(segments_generator)
+        return {
+            "text": " ".join(segment.text for segment in segments),
+            "segments": segments,
+            "language": info.language,
+        }
+
+
+def recognize(recognizer, audio_data: AudioData) -> str:
+    import torch
+    from faster_whisper import WhisperModel
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+
+    model = WhisperModel("base", device=device)
+    whisper_recognizer = WhisperCompatibleRecognizer(
+        TranscribableAdapter(model)
+    )
+    return whisper_recognizer.recognize(audio_data)
diff --git a/speech_recognition/recognizers/whisper_local/whisper.py b/speech_recognition/recognizers/whisper_local/whisper.py
index 1c8299bb..7e05b94a 100644
--- a/speech_recognition/recognizers/whisper_local/whisper.py
+++ b/speech_recognition/recognizers/whisper_local/whisper.py
@@ -10,6 +10,7 @@
 if TYPE_CHECKING:
     import torch
     from typing_extensions import Unpack
+    from whisper import Whisper
 
 
 class LoadModelOptionalParameters(TypedDict, total=False):
@@ -52,6 +53,19 @@ class TranscribeOutput(TypedDict):
     language: str
 
 
+class TranscribableAdapter:
+    def __init__(self, model: Whisper) -> None:
+        self.model = model
+
+    def transcribe(self, audio_array, **kwargs) -> TranscribeOutput:
+        if "fp16" not in kwargs:
+            import torch
+
+            kwargs["fp16"] = torch.cuda.is_available()
+
+        return self.model.transcribe(audio_array, **kwargs)
+
+
 def recognize(
     recognizer,
     audio_data: AudioData,
@@ -80,7 +94,9 @@ def recognize(
     import whisper
 
     whisper_model = whisper.load_model(model, **load_options or {})
-    whisper_recognizer = WhisperCompatibleRecognizer(whisper_model)
+    whisper_recognizer = WhisperCompatibleRecognizer(
+        TranscribableAdapter(whisper_model)
+    )
     return whisper_recognizer.recognize(
         audio_data, show_dict=show_dict, **transcribe_options
     )
diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
new file mode 100644
index 00000000..4d4f04fd
--- /dev/null
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -0,0 +1,75 @@
+import sys
+from unittest.mock import ANY, MagicMock, patch
+
+import numpy as np
+import pytest
+from faster_whisper.transcribe import (
+    Segment,
+    TranscriptionInfo,
+    TranscriptionOptions,
+    VadOptions,
+)
+
+from speech_recognition import Recognizer
+from speech_recognition.audio import AudioData
+from speech_recognition.recognizers.whisper_local.faster_whisper import (
+    recognize,
+)
+
+
+@pytest.mark.skipif(
+    sys.version_info >= (3, 13), reason="skip on Python 3.13 or later"
+)
+@patch("soundfile.read")
+@patch("torch.cuda.is_available", return_value=False)
+@patch("faster_whisper.WhisperModel")
+def test_transcribe_with_default_parameters(
+    WhisperModel, is_available, sf_read
+):
+    def segments():
+        yield Segment(
+            id=1,
+            seek=0,
+            start=0.0,
+            end=2.64,
+            text=" 1, 2, 3",
+            tokens=[50364, 502, 11, 568, 11, 805, 50496],
+            avg_logprob=-0.5378808751702309,
+            compression_ratio=0.4666666666666667,
+            no_speech_prob=0.17316274344921112,
+            words=None,
+            temperature=0.0,
+        )
+
+    info = TranscriptionInfo(
+        language="en",
+        language_probability=0.9314374923706055,
+        duration=2.7449375,
+        duration_after_vad=2.7449375,
+        all_language_probs=[("en", 0.9314374923706055)],
+        transcription_options=MagicMock(spec=TranscriptionOptions),
+        vad_options=MagicMock(spec=VadOptions),
+    )
+
+    whisper_model = WhisperModel.return_value
+    whisper_model.transcribe.return_value = segments(), info
+
+    audio_data = MagicMock(spec=AudioData)
+    audio_data.get_wav_data.return_value = b"audio data"
+
+    audio_array = MagicMock(spec=np.ndarray)
+    dummy_sampling_rate = 99_999
+    sf_read.return_value = (audio_array, dummy_sampling_rate)
+
+    actual = recognize(MagicMock(spec=Recognizer), audio_data)
+
+    assert actual == " 1, 2, 3"
+    is_available.assert_called_once_with()
+    WhisperModel.assert_called_once_with("base", device="cpu")
+    audio_data.get_wav_data.assert_called_once_with(convert_rate=16_000)
+    sf_read.assert_called_once_with(ANY)
+    assert sf_read.call_args[0][0].read() == b"audio data"
+    audio_array.astype.assert_called_once_with(np.float32)
+    whisper_model.transcribe.assert_called_once_with(
+        audio_array.astype.return_value
+    )

From cbca8b994bbed0817a71447857e03d16743ab5ad Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:00:29 +0900
Subject: [PATCH 02/22] [refactor] Reduce mock in whisper tests

---
 .../recognizers/whisper_local/test_whisper.py | 28 +++++++++----------
 1 file changed, 14 insertions(+), 14 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_whisper.py b/tests/recognizers/whisper_local/test_whisper.py
index be9c305b..5fbb6f4c 100644
--- a/tests/recognizers/whisper_local/test_whisper.py
+++ b/tests/recognizers/whisper_local/test_whisper.py
@@ -1,47 +1,45 @@
 import sys
 from unittest import TestCase, skipIf
-from unittest.mock import MagicMock, patch
+from unittest.mock import ANY, MagicMock, patch
+
+import numpy as np
 
 from speech_recognition import AudioData, Recognizer
 from speech_recognition.recognizers.whisper_local.whisper import recognize
 
 
 @skipIf(sys.version_info >= (3, 13), "skip on Python 3.13")
-@patch("speech_recognition.recognizers.whisper_local.base.io.BytesIO")
 @patch("soundfile.read")
 @patch("torch.cuda.is_available")
 @patch("whisper.load_model")
 class RecognizeWhisperTestCase(TestCase):
-    def test_default_parameters(
-        self, load_model, is_available, sf_read, BytesIO
-    ):
-        import numpy as np
-
+    def test_default_parameters(self, load_model, is_available, sf_read):
         whisper_model = load_model.return_value
         whisper_model.transcribe.return_value = {
             "text": "Transcription by Whisper model",
             "language": "en",
             # Omit "segments"
         }
-        audio_array = MagicMock()
+        audio_array = MagicMock(spec=np.ndarray)
         dummy_sampling_rate = 99_999
         sf_read.return_value = (audio_array, dummy_sampling_rate)
 
         audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b"wav_data"
         actual = recognize(MagicMock(spec=Recognizer), audio_data)
 
         assert actual == "Transcription by Whisper model"
         load_model.assert_called_once_with("base")
         audio_data.get_wav_data.assert_called_once_with(convert_rate=16000)
-        BytesIO.assert_called_once_with(audio_data.get_wav_data.return_value)
-        sf_read.assert_called_once_with(BytesIO.return_value)
+        sf_read.assert_called_once_with(ANY)
+        assert sf_read.call_args[0][0].read() == b"wav_data"
         audio_array.astype.assert_called_once_with(np.float32)
         whisper_model.transcribe.assert_called_once_with(
             audio_array.astype.return_value,
             fp16=is_available.return_value,
         )
 
-    def test_return_as_dict(self, load_model, is_available, sf_read, BytesIO):
+    def test_return_as_dict(self, load_model, is_available, sf_read):
         whisper_model = load_model.return_value
         whisper_model.transcribe.return_value = {
             "text": " 1, 2, 3",
@@ -61,11 +59,12 @@ def test_return_as_dict(self, load_model, is_available, sf_read, BytesIO):
             ],
             "language": "en",
         }
-        audio_array = MagicMock()
+        audio_array = MagicMock(spec=np.ndarray)
         dummy_sampling_rate = 99_999
         sf_read.return_value = (audio_array, dummy_sampling_rate)
 
         audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b""
         actual = recognize(
             MagicMock(spec=Recognizer), audio_data, show_dict=True
         )
@@ -91,13 +90,14 @@ def test_return_as_dict(self, load_model, is_available, sf_read, BytesIO):
 
         assert actual == expected
 
-    def test_pass_parameters(self, load_model, is_available, sf_read, BytesIO):
+    def test_pass_parameters(self, load_model, is_available, sf_read):
         whisper_model = load_model.return_value
-        audio_array = MagicMock()
+        audio_array = MagicMock(spec=np.ndarray)
         dummy_sampling_rate = 99_999
         sf_read.return_value = (audio_array, dummy_sampling_rate)
 
         audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b""
         _ = recognize(
             MagicMock(spec=Recognizer),
             audio_data,

From 7df12536e7f940ffe6993a7123bb422b2fb75016 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:06:58 +0900
Subject: [PATCH 03/22] [refactor] Test class to add other case easier

---
 .../whisper_local/test_faster_whisper.py      | 89 +++++++++----------
 1 file changed, 44 insertions(+), 45 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 4d4f04fd..ddd98352 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -21,55 +21,54 @@
     sys.version_info >= (3, 13), reason="skip on Python 3.13 or later"
 )
 @patch("soundfile.read")
-@patch("torch.cuda.is_available", return_value=False)
 @patch("faster_whisper.WhisperModel")
-def test_transcribe_with_default_parameters(
-    WhisperModel, is_available, sf_read
-):
-    def segments():
-        yield Segment(
-            id=1,
-            seek=0,
-            start=0.0,
-            end=2.64,
-            text=" 1, 2, 3",
-            tokens=[50364, 502, 11, 568, 11, 805, 50496],
-            avg_logprob=-0.5378808751702309,
-            compression_ratio=0.4666666666666667,
-            no_speech_prob=0.17316274344921112,
-            words=None,
-            temperature=0.0,
-        )
+class TestTranscribe:
+    @patch("torch.cuda.is_available", return_value=False)
+    def test_default_parameters(self, is_available, WhisperModel, sf_read):
+        def segments():
+            yield Segment(
+                id=1,
+                seek=0,
+                start=0.0,
+                end=2.64,
+                text=" 1, 2, 3",
+                tokens=[50364, 502, 11, 568, 11, 805, 50496],
+                avg_logprob=-0.5378808751702309,
+                compression_ratio=0.4666666666666667,
+                no_speech_prob=0.17316274344921112,
+                words=None,
+                temperature=0.0,
+            )
 
-    info = TranscriptionInfo(
-        language="en",
-        language_probability=0.9314374923706055,
-        duration=2.7449375,
-        duration_after_vad=2.7449375,
-        all_language_probs=[("en", 0.9314374923706055)],
-        transcription_options=MagicMock(spec=TranscriptionOptions),
-        vad_options=MagicMock(spec=VadOptions),
-    )
+        info = TranscriptionInfo(
+            language="en",
+            language_probability=0.9314374923706055,
+            duration=2.7449375,
+            duration_after_vad=2.7449375,
+            all_language_probs=[("en", 0.9314374923706055)],
+            transcription_options=MagicMock(spec=TranscriptionOptions),
+            vad_options=MagicMock(spec=VadOptions),
+        )
 
-    whisper_model = WhisperModel.return_value
-    whisper_model.transcribe.return_value = segments(), info
+        whisper_model = WhisperModel.return_value
+        whisper_model.transcribe.return_value = segments(), info
 
-    audio_data = MagicMock(spec=AudioData)
-    audio_data.get_wav_data.return_value = b"audio data"
+        audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b"audio data"
 
-    audio_array = MagicMock(spec=np.ndarray)
-    dummy_sampling_rate = 99_999
-    sf_read.return_value = (audio_array, dummy_sampling_rate)
+        audio_array = MagicMock(spec=np.ndarray)
+        dummy_sampling_rate = 99_999
+        sf_read.return_value = (audio_array, dummy_sampling_rate)
 
-    actual = recognize(MagicMock(spec=Recognizer), audio_data)
+        actual = recognize(MagicMock(spec=Recognizer), audio_data)
 
-    assert actual == " 1, 2, 3"
-    is_available.assert_called_once_with()
-    WhisperModel.assert_called_once_with("base", device="cpu")
-    audio_data.get_wav_data.assert_called_once_with(convert_rate=16_000)
-    sf_read.assert_called_once_with(ANY)
-    assert sf_read.call_args[0][0].read() == b"audio data"
-    audio_array.astype.assert_called_once_with(np.float32)
-    whisper_model.transcribe.assert_called_once_with(
-        audio_array.astype.return_value
-    )
+        assert actual == " 1, 2, 3"
+        is_available.assert_called_once_with()
+        WhisperModel.assert_called_once_with("base", device="cpu")
+        audio_data.get_wav_data.assert_called_once_with(convert_rate=16_000)
+        sf_read.assert_called_once_with(ANY)
+        assert sf_read.call_args[0][0].read() == b"audio data"
+        audio_array.astype.assert_called_once_with(np.float32)
+        whisper_model.transcribe.assert_called_once_with(
+            audio_array.astype.return_value
+        )

From 36b046493205b44db2fad9f6641f4186932a4bb6 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:22:56 +0900
Subject: [PATCH 04/22] [test] GPU available case

---
 .../whisper_local/test_faster_whisper.py      | 23 +++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index ddd98352..e2ce46e1 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -72,3 +72,26 @@ def segments():
         whisper_model.transcribe.assert_called_once_with(
             audio_array.astype.return_value
         )
+
+    @patch("torch.cuda.is_available", return_value=True)
+    def test_gpu_available(self, is_available, WhisperModel, sf_read):
+        def segments_generator():
+            mocked_segment = MagicMock(spec=Segment(*[None] * 11))
+            mocked_segment.text = ""
+            yield mocked_segment
+
+        WhisperModel.return_value.transcribe.return_value = (
+            segments_generator(),
+            MagicMock(spec=TranscriptionInfo(*[None] * 7)),
+        )
+
+        audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b""
+
+        audio_array = MagicMock(spec=np.ndarray)
+        dummy_sampling_rate = 99_999
+        sf_read.return_value = (audio_array, dummy_sampling_rate)
+
+        _ = recognize(MagicMock(spec=Recognizer), audio_data)
+
+        WhisperModel.assert_called_once_with("base", device="cuda")

From 3dbbe173c430b88d4be40ea997643e4556c048a8 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:39:45 +0900
Subject: [PATCH 05/22] [feat] Specify transcribe's parameters

---
 .../whisper_local/faster_whisper.py           | 19 +++++++---
 .../whisper_local/test_faster_whisper.py      | 38 +++++++++++++++++++
 2 files changed, 52 insertions(+), 5 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index 58d4b860..c512bccf 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -8,6 +8,7 @@
 )
 
 if TYPE_CHECKING:
+    import numpy as np
     from faster_whisper import WhisperModel
     from faster_whisper.transcribe import Segment
 
@@ -22,8 +23,10 @@ class TranscribableAdapter:
     def __init__(self, model: WhisperModel) -> None:
         self.model = model
 
-    def transcribe(self, audio_array, **kwargs) -> TranscribeOutput:
-        segments_generator, info = self.model.transcribe(audio_array)
+    def transcribe(
+        self, audio_array: np.ndarray, **kwargs
+    ) -> TranscribeOutput:
+        segments_generator, info = self.model.transcribe(audio_array, **kwargs)
         segments = list(segments_generator)
         return {
             "text": " ".join(segment.text for segment in segments),
@@ -32,14 +35,20 @@ def transcribe(self, audio_array, **kwargs) -> TranscribeOutput:
         }
 
 
-def recognize(recognizer, audio_data: AudioData) -> str:
+def recognize(
+    recognizer,
+    audio_data: AudioData,
+    model: str = "base",
+    show_dict: bool = False,
+    **transcribe_options,
+) -> str:
     import torch
     from faster_whisper import WhisperModel
 
     device = "cuda" if torch.cuda.is_available() else "cpu"
 
-    model = WhisperModel("base", device=device)
+    model = WhisperModel(model, device=device)
     whisper_recognizer = WhisperCompatibleRecognizer(
         TranscribableAdapter(model)
     )
-    return whisper_recognizer.recognize(audio_data)
+    return whisper_recognizer.recognize(audio_data, **transcribe_options)
diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index e2ce46e1..d47ff702 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -95,3 +95,41 @@ def segments_generator():
         _ = recognize(MagicMock(spec=Recognizer), audio_data)
 
         WhisperModel.assert_called_once_with("base", device="cuda")
+
+    @patch("torch.cuda.is_available", return_value=False)
+    def test_pass_parameters(self, is_available, WhisperModel, sf_read):
+        def segments_generator():
+            mocked_segment = MagicMock(spec=Segment(*[None] * 11))
+            mocked_segment.text = ""
+            yield mocked_segment
+
+        whisper_model = WhisperModel.return_value
+        whisper_model.transcribe.return_value = (
+            segments_generator(),
+            MagicMock(spec=TranscriptionInfo(*[None] * 7)),
+        )
+
+        audio_data = MagicMock(spec=AudioData)
+        audio_data.get_wav_data.return_value = b""
+
+        audio_array = MagicMock(spec=np.ndarray)
+        dummy_sampling_rate = 99_999
+        sf_read.return_value = (audio_array, dummy_sampling_rate)
+
+        _ = recognize(
+            MagicMock(spec=Recognizer),
+            audio_data,
+            model="small",
+            show_dict=True,
+            language="fr",
+            task="translate",
+            beam_size=5,
+        )
+
+        WhisperModel.assert_called_once_with("small", device="cpu")
+        whisper_model.transcribe.assert_called_once_with(
+            audio_array.astype.return_value,
+            language="fr",
+            task="translate",
+            beam_size=5,
+        )

From 2ee6044fa57aba46b9f808c75070bbb3ba9c4b59 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:43:58 +0900
Subject: [PATCH 06/22] [refactor] Extract AudioData fixture

---
 .../whisper_local/test_faster_whisper.py      | 26 ++++++++++++-------
 1 file changed, 16 insertions(+), 10 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index d47ff702..3bcb7a6b 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -17,6 +17,13 @@
 )
 
 
+@pytest.fixture
+def audio_data() -> AudioData:
+    audio = MagicMock(spec=AudioData)
+    audio.get_wav_data.return_value = b""
+    return audio
+
+
 @pytest.mark.skipif(
     sys.version_info >= (3, 13), reason="skip on Python 3.13 or later"
 )
@@ -24,7 +31,9 @@
 @patch("faster_whisper.WhisperModel")
 class TestTranscribe:
     @patch("torch.cuda.is_available", return_value=False)
-    def test_default_parameters(self, is_available, WhisperModel, sf_read):
+    def test_default_parameters(
+        self, is_available, WhisperModel, sf_read, audio_data
+    ):
         def segments():
             yield Segment(
                 id=1,
@@ -53,7 +62,6 @@ def segments():
         whisper_model = WhisperModel.return_value
         whisper_model.transcribe.return_value = segments(), info
 
-        audio_data = MagicMock(spec=AudioData)
         audio_data.get_wav_data.return_value = b"audio data"
 
         audio_array = MagicMock(spec=np.ndarray)
@@ -74,7 +82,9 @@ def segments():
         )
 
     @patch("torch.cuda.is_available", return_value=True)
-    def test_gpu_available(self, is_available, WhisperModel, sf_read):
+    def test_gpu_available(
+        self, is_available, WhisperModel, sf_read, audio_data
+    ):
         def segments_generator():
             mocked_segment = MagicMock(spec=Segment(*[None] * 11))
             mocked_segment.text = ""
@@ -85,9 +95,6 @@ def segments_generator():
             MagicMock(spec=TranscriptionInfo(*[None] * 7)),
         )
 
-        audio_data = MagicMock(spec=AudioData)
-        audio_data.get_wav_data.return_value = b""
-
         audio_array = MagicMock(spec=np.ndarray)
         dummy_sampling_rate = 99_999
         sf_read.return_value = (audio_array, dummy_sampling_rate)
@@ -97,7 +104,9 @@ def segments_generator():
         WhisperModel.assert_called_once_with("base", device="cuda")
 
     @patch("torch.cuda.is_available", return_value=False)
-    def test_pass_parameters(self, is_available, WhisperModel, sf_read):
+    def test_pass_parameters(
+        self, is_available, WhisperModel, sf_read, audio_data
+    ):
         def segments_generator():
             mocked_segment = MagicMock(spec=Segment(*[None] * 11))
             mocked_segment.text = ""
@@ -109,9 +118,6 @@ def segments_generator():
             MagicMock(spec=TranscriptionInfo(*[None] * 7)),
         )
 
-        audio_data = MagicMock(spec=AudioData)
-        audio_data.get_wav_data.return_value = b""
-
         audio_array = MagicMock(spec=np.ndarray)
         dummy_sampling_rate = 99_999
         sf_read.return_value = (audio_array, dummy_sampling_rate)

From 4dd2985bc1e9c4ffbc40c6ad2160456d3772b337 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:54:37 +0900
Subject: [PATCH 07/22] [refactor] Patch sf.read as fixture

---
 .../whisper_local/test_faster_whisper.py      | 35 ++++++++++---------
 1 file changed, 18 insertions(+), 17 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 3bcb7a6b..87a3732d 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -1,3 +1,4 @@
+from collections.abc import Generator
 import sys
 from unittest.mock import ANY, MagicMock, patch
 
@@ -24,16 +25,27 @@ def audio_data() -> AudioData:
     return audio
 
 
+@pytest.fixture
+def soundfile_read() -> Generator[tuple[MagicMock, np.ndarray], None, None]:
+    audio_array = MagicMock(spec=np.ndarray)
+    dummy_sampling_rate = 99_999
+
+    with patch("soundfile.read") as mock_read:
+        mock_read.return_value = (audio_array, dummy_sampling_rate)
+        yield mock_read, audio_array
+
+
 @pytest.mark.skipif(
     sys.version_info >= (3, 13), reason="skip on Python 3.13 or later"
 )
-@patch("soundfile.read")
 @patch("faster_whisper.WhisperModel")
 class TestTranscribe:
     @patch("torch.cuda.is_available", return_value=False)
     def test_default_parameters(
-        self, is_available, WhisperModel, sf_read, audio_data
+        self, is_available, WhisperModel, audio_data, soundfile_read
     ):
+        sf_read, audio_array = soundfile_read
+
         def segments():
             yield Segment(
                 id=1,
@@ -63,11 +75,6 @@ def segments():
         whisper_model.transcribe.return_value = segments(), info
 
         audio_data.get_wav_data.return_value = b"audio data"
-
-        audio_array = MagicMock(spec=np.ndarray)
-        dummy_sampling_rate = 99_999
-        sf_read.return_value = (audio_array, dummy_sampling_rate)
-
         actual = recognize(MagicMock(spec=Recognizer), audio_data)
 
         assert actual == " 1, 2, 3"
@@ -83,7 +90,7 @@ def segments():
 
     @patch("torch.cuda.is_available", return_value=True)
     def test_gpu_available(
-        self, is_available, WhisperModel, sf_read, audio_data
+        self, is_available, WhisperModel, audio_data, soundfile_read
     ):
         def segments_generator():
             mocked_segment = MagicMock(spec=Segment(*[None] * 11))
@@ -95,18 +102,16 @@ def segments_generator():
             MagicMock(spec=TranscriptionInfo(*[None] * 7)),
         )
 
-        audio_array = MagicMock(spec=np.ndarray)
-        dummy_sampling_rate = 99_999
-        sf_read.return_value = (audio_array, dummy_sampling_rate)
-
         _ = recognize(MagicMock(spec=Recognizer), audio_data)
 
         WhisperModel.assert_called_once_with("base", device="cuda")
 
     @patch("torch.cuda.is_available", return_value=False)
     def test_pass_parameters(
-        self, is_available, WhisperModel, sf_read, audio_data
+        self, is_available, WhisperModel, audio_data, soundfile_read
     ):
+        _, audio_array = soundfile_read
+
         def segments_generator():
             mocked_segment = MagicMock(spec=Segment(*[None] * 11))
             mocked_segment.text = ""
@@ -118,10 +123,6 @@ def segments_generator():
             MagicMock(spec=TranscriptionInfo(*[None] * 7)),
         )
 
-        audio_array = MagicMock(spec=np.ndarray)
-        dummy_sampling_rate = 99_999
-        sf_read.return_value = (audio_array, dummy_sampling_rate)
-
         _ = recognize(
             MagicMock(spec=Recognizer),
             audio_data,

From 8b2d91c302dbfe5e987268d93ce5c8c618564f7e Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 14:59:48 +0900
Subject: [PATCH 08/22] [refactor] Extract mock creation into fixtures

---
 .../whisper_local/test_faster_whisper.py      | 44 ++++++++++++++-----
 1 file changed, 32 insertions(+), 12 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 87a3732d..750ddd29 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -1,5 +1,5 @@
-from collections.abc import Generator
 import sys
+from collections.abc import Generator
 from unittest.mock import ANY, MagicMock, patch
 
 import numpy as np
@@ -25,6 +25,18 @@ def audio_data() -> AudioData:
     return audio
 
 
+@pytest.fixture
+def segment() -> Segment:
+    mocked_segment = MagicMock(spec=Segment(*[None] * 11))
+    mocked_segment.text = ""
+    return mocked_segment
+
+
+@pytest.fixture
+def transcription_info() -> TranscriptionInfo:
+    return MagicMock(spec=TranscriptionInfo(*[None] * 7))
+
+
 @pytest.fixture
 def soundfile_read() -> Generator[tuple[MagicMock, np.ndarray], None, None]:
     audio_array = MagicMock(spec=np.ndarray)
@@ -66,7 +78,7 @@ def segments():
             language_probability=0.9314374923706055,
             duration=2.7449375,
             duration_after_vad=2.7449375,
-            all_language_probs=[("en", 0.9314374923706055)],
+            all_language_probs=[("en", 0.9314374923706055)],  # Omitted
             transcription_options=MagicMock(spec=TranscriptionOptions),
             vad_options=MagicMock(spec=VadOptions),
         )
@@ -90,16 +102,20 @@ def segments():
 
     @patch("torch.cuda.is_available", return_value=True)
     def test_gpu_available(
-        self, is_available, WhisperModel, audio_data, soundfile_read
+        self,
+        is_available,
+        WhisperModel,
+        audio_data,
+        segment,
+        transcription_info,
+        soundfile_read,
     ):
         def segments_generator():
-            mocked_segment = MagicMock(spec=Segment(*[None] * 11))
-            mocked_segment.text = ""
-            yield mocked_segment
+            yield segment
 
         WhisperModel.return_value.transcribe.return_value = (
             segments_generator(),
-            MagicMock(spec=TranscriptionInfo(*[None] * 7)),
+            transcription_info,
         )
 
         _ = recognize(MagicMock(spec=Recognizer), audio_data)
@@ -108,19 +124,23 @@ def segments_generator():
 
     @patch("torch.cuda.is_available", return_value=False)
     def test_pass_parameters(
-        self, is_available, WhisperModel, audio_data, soundfile_read
+        self,
+        is_available,
+        WhisperModel,
+        audio_data,
+        segment,
+        transcription_info,
+        soundfile_read,
     ):
         _, audio_array = soundfile_read
 
         def segments_generator():
-            mocked_segment = MagicMock(spec=Segment(*[None] * 11))
-            mocked_segment.text = ""
-            yield mocked_segment
+            yield segment
 
         whisper_model = WhisperModel.return_value
         whisper_model.transcribe.return_value = (
             segments_generator(),
-            MagicMock(spec=TranscriptionInfo(*[None] * 7)),
+            transcription_info,
         )
 
         _ = recognize(

From d502a2e7a0efde6153203d5f74efa73c41e3c1b9 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:00:24 +0900
Subject: [PATCH 09/22] [docs] Add type hints

---
 speech_recognition/recognizers/whisper_local/base.py    | 9 +++++++--
 speech_recognition/recognizers/whisper_local/whisper.py | 5 ++++-
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/base.py b/speech_recognition/recognizers/whisper_local/base.py
index 12e04e65..ad6ee101 100644
--- a/speech_recognition/recognizers/whisper_local/base.py
+++ b/speech_recognition/recognizers/whisper_local/base.py
@@ -1,13 +1,18 @@
 from __future__ import annotations
 
 import io
-from typing import Any, Protocol
+from typing import TYPE_CHECKING, Any, Protocol
 
 from speech_recognition.audio import AudioData
 
+if TYPE_CHECKING:
+    import numpy as np
+
 
 class Transcribable(Protocol):
-    def transcribe(self, audio_array, **kwargs) -> str | dict[str, Any]:
+    def transcribe(
+        self, audio_array: np.ndarray, **kwargs
+    ) -> str | dict[str, Any]:
         pass
 
 
diff --git a/speech_recognition/recognizers/whisper_local/whisper.py b/speech_recognition/recognizers/whisper_local/whisper.py
index 7e05b94a..c4642e22 100644
--- a/speech_recognition/recognizers/whisper_local/whisper.py
+++ b/speech_recognition/recognizers/whisper_local/whisper.py
@@ -8,6 +8,7 @@
 )
 
 if TYPE_CHECKING:
+    import numpy as np
     import torch
     from typing_extensions import Unpack
     from whisper import Whisper
@@ -57,7 +58,9 @@ class TranscribableAdapter:
     def __init__(self, model: Whisper) -> None:
         self.model = model
 
-    def transcribe(self, audio_array, **kwargs) -> TranscribeOutput:
+    def transcribe(
+        self, audio_array: np.ndarray, **kwargs
+    ) -> TranscribeOutput:
         if "fp16" not in kwargs:
             import torch
 

From 5da7fdffa87c5c338c5385ba74057574d6dc4563 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:03:29 +0900
Subject: [PATCH 10/22] [chore] faster-whisper extra

---
 .github/workflows/unittests.yml | 4 ++--
 setup.cfg                       | 2 ++
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/unittests.yml b/.github/workflows/unittests.yml
index 170b78f8..9b460c2f 100644
--- a/.github/workflows/unittests.yml
+++ b/.github/workflows/unittests.yml
@@ -44,7 +44,7 @@ jobs:
       - name: Install Python dependencies (Ubuntu, <=3.12)
         if: matrix.os == 'ubuntu-latest' && matrix.python-version != '3.13'
         run: |
-          python -m pip install .[dev,audio,pocketsphinx,google-cloud,whisper-local,openai,groq]
+          python -m pip install .[dev,audio,pocketsphinx,google-cloud,whisper-local,faster-whisper,openai,groq]
       - name: Install Python dependencies (Ubuntu, 3.13)
         if: matrix.os == 'ubuntu-latest' && matrix.python-version == '3.13'
         run: |
@@ -53,7 +53,7 @@ jobs:
       - name: Install Python dependencies (Windows)
         if: matrix.os == 'windows-latest'
         run: |
-          python -m pip install .[dev,whisper-local,google-cloud,openai,groq]
+          python -m pip install .[dev,whisper-local,faster-whisper,google-cloud,openai,groq]
       - name: Test with unittest
         run: |
           pytest --doctest-modules -v speech_recognition/recognizers/ tests/
diff --git a/setup.cfg b/setup.cfg
index 9d1def40..f3848d4d 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -12,6 +12,8 @@ google-cloud =
 whisper-local =
     openai-whisper
     soundfile
+faster-whisper =
+    faster-whisper
 openai =
     openai
     httpx < 0.28

From 5c340845fc4eff6de2ef46a405eba4fd6979a1ad Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:09:31 +0900
Subject: [PATCH 11/22] [docs] Add parameters' type hint

---
 .../recognizers/whisper_local/faster_whisper.py     | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index c512bccf..a4fe2105 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import TYPE_CHECKING, TypedDict
+from typing import TYPE_CHECKING, Literal, TypedDict
 
 from speech_recognition.audio import AudioData
 from speech_recognition.recognizers.whisper_local.base import (
@@ -11,6 +11,7 @@
     import numpy as np
     from faster_whisper import WhisperModel
     from faster_whisper.transcribe import Segment
+    from typing_extensions import Unpack
 
 
 class TranscribeOutput(TypedDict):
@@ -35,12 +36,20 @@ def transcribe(
         }
 
 
+class TranscribeOptionalParameters(TypedDict, total=False):
+    # https://github.com/SYSTRAN/faster-whisper/blob/v1.1.0/faster_whisper/transcribe.py#L692
+    language: str
+    task: Literal["transcribe", "translate"]
+    beam_size: int
+    # TODO Add others
+
+
 def recognize(
     recognizer,
     audio_data: AudioData,
     model: str = "base",
     show_dict: bool = False,
-    **transcribe_options,
+    **transcribe_options: Unpack[TranscribeOptionalParameters],
 ) -> str:
     import torch
     from faster_whisper import WhisperModel

From 5f55e920c6bf102b45492e903c3b7847bfe1904b Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:14:44 +0900
Subject: [PATCH 12/22] [bugfix] show_dict=True

---
 .../whisper_local/faster_whisper.py           |  4 +-
 .../whisper_local/test_faster_whisper.py      | 59 +++++++++++++++++++
 2 files changed, 62 insertions(+), 1 deletion(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index a4fe2105..1593d4e3 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -60,4 +60,6 @@ def recognize(
     whisper_recognizer = WhisperCompatibleRecognizer(
         TranscribableAdapter(model)
     )
-    return whisper_recognizer.recognize(audio_data, **transcribe_options)
+    return whisper_recognizer.recognize(
+        audio_data, show_dict=show_dict, **transcribe_options
+    )
diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 750ddd29..06a6b0e6 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -122,6 +122,65 @@ def segments_generator():
 
         WhisperModel.assert_called_once_with("base", device="cuda")
 
+    @patch("torch.cuda.is_available", return_value=False)
+    def test_show_dict(
+        self, is_available, WhisperModel, audio_data, soundfile_read
+    ):
+        sf_read, audio_array = soundfile_read
+
+        def segments():
+            yield Segment(
+                id=1,
+                seek=0,
+                start=0.0,
+                end=2.64,
+                text=" 1, 2, 3",
+                tokens=[50364, 502, 11, 568, 11, 805, 50496],
+                avg_logprob=-0.5378808751702309,
+                compression_ratio=0.4666666666666667,
+                no_speech_prob=0.17316274344921112,
+                words=None,
+                temperature=0.0,
+            )
+
+        info = TranscriptionInfo(
+            language="en",
+            language_probability=0.9314374923706055,
+            duration=2.7449375,
+            duration_after_vad=2.7449375,
+            all_language_probs=[("en", 0.9314374923706055)],  # Omitted
+            transcription_options=MagicMock(spec=TranscriptionOptions),
+            vad_options=MagicMock(spec=VadOptions),
+        )
+
+        whisper_model = WhisperModel.return_value
+        whisper_model.transcribe.return_value = segments(), info
+
+        actual = recognize(
+            MagicMock(spec=Recognizer), audio_data, show_dict=True
+        )
+
+        expected = {
+            "text": " 1, 2, 3",
+            "language": "en",
+            "segments": [
+                Segment(
+                    id=1,
+                    seek=0,
+                    start=0.0,
+                    end=2.64,
+                    text=" 1, 2, 3",
+                    tokens=[50364, 502, 11, 568, 11, 805, 50496],
+                    avg_logprob=-0.5378808751702309,
+                    compression_ratio=0.4666666666666667,
+                    no_speech_prob=0.17316274344921112,
+                    words=None,
+                    temperature=0.0,
+                )
+            ],
+        }
+        assert actual == expected
+
     @patch("torch.cuda.is_available", return_value=False)
     def test_pass_parameters(
         self,

From a422db06e00bac1bf8ba3879a615f5660593f898 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:28:17 +0900
Subject: [PATCH 13/22] [docs] Add docstring

---
 .../whisper_local/faster_whisper.py            | 18 ++++++++++++++++++
 .../recognizers/whisper_local/whisper.py       |  7 +++++--
 2 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index 1593d4e3..ffc858b3 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -51,6 +51,24 @@ def recognize(
     show_dict: bool = False,
     **transcribe_options: Unpack[TranscribeOptionalParameters],
 ) -> str:
+    """Performs speech recognition on ``audio_data`` (an ``AudioData`` instance), using Whisper.
+
+    Pick ``model`` size (Same as Whisper).
+
+    If ``show_dict`` is true, returns the detailed response from Whisper, including the detected language. Otherwise returns only the transcription.
+
+    You can specify:
+
+        * ``language``: recognition language, an uncapitalized 2 letters language name like "en" or "fr".
+
+            * If not set, Faster Whisper will automatically detect the language.
+
+        * ``task``
+
+            * If you want transcribe + **translate** to english, set ``task="translate"``.
+
+    Other values are passed directly to whisper. See https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/transcribe.py for all options.
+    """
     import torch
     from faster_whisper import WhisperModel
 
diff --git a/speech_recognition/recognizers/whisper_local/whisper.py b/speech_recognition/recognizers/whisper_local/whisper.py
index c4642e22..622ee071 100644
--- a/speech_recognition/recognizers/whisper_local/whisper.py
+++ b/speech_recognition/recognizers/whisper_local/whisper.py
@@ -87,11 +87,14 @@ def recognize(
     You can specify:
 
         * ``language``: recognition language, an uncapitalized full language name like "english" or "chinese". See the full language list at https://github.com/openai/whisper/blob/main/whisper/tokenizer.py
+
+            * If not set, Whisper will automatically detect the language.
+
         * ``task``
 
-            * If you want transcribe + **translate**, set ``task="translate"``.
+            * If you want transcribe + **translate** to english, set ``task="translate"``.
 
-    Other values are passed directly to whisper. See https://github.com/openai/whisper/blob/main/whisper/transcribe.py for all options
+    Other values are passed directly to whisper. See https://github.com/openai/whisper/blob/main/whisper/transcribe.py for all options.
     """
 
     import whisper

From ec8d05f9afb108d04f41e32ab82d1e99e4b1e3f2 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:31:07 +0900
Subject: [PATCH 14/22] [docs] Add faster_whisper in reference

---
 reference/library-reference.rst | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/reference/library-reference.rst b/reference/library-reference.rst
index dd5c2169..32ee973a 100644
--- a/reference/library-reference.rst
+++ b/reference/library-reference.rst
@@ -290,11 +290,16 @@ Returns the most likely transcription if ``show_all`` is false (the default). Ot
 
 Raises a ``speech_recognition.UnknownValueError`` exception if the speech is unintelligible. Raises a ``speech_recognition.RequestError`` exception if the speech recognition operation failed, if the key isn't valid, or if there is no internet connection.
 
-``recognizer_instance.recognize_whisper(audio_data: AudioData, model: str="base", show_dict: bool=False, load_options=None, **transcribe_options):``
-----------------------------------------------------------------------------------------------------------------------------------------------------
+``recognizer_instance.recognize_whisper(audio_data: AudioData, model: str="base", show_dict: bool=False, load_options=None, **transcribe_options)``
+---------------------------------------------------------------------------------------------------------------------------------------------------
 
 .. autofunction:: speech_recognition.recognizers.whisper_local.whisper.recognize
 
+``recognizer_instance.recognize_faster_whisper(audio_data: AudioData, model: str="base", show_dict: bool=False, **transcribe_options)``
+---------------------------------------------------------------------------------------------------------------------------------------
+
+.. autofunction:: speech_recognition.recognizers.whisper_local.faster_whisper.recognize
+
 ``recognizer_instance.recognize_openai(audio_data: AudioData, model = "whisper-1", **kwargs)``
 ----------------------------------------------------------------------------------------------
 

From 4f2506aef5fb5205ef7c370f4ae286659c161a6e Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:31:53 +0900
Subject: [PATCH 15/22] [feat] Users can call
 Recognizer.recognize_faster_whisper()

---
 speech_recognition/__init__.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/speech_recognition/__init__.py b/speech_recognition/__init__.py
index 96d6d74f..140d8cba 100644
--- a/speech_recognition/__init__.py
+++ b/speech_recognition/__init__.py
@@ -1392,13 +1392,14 @@ def flush(self, *args, **kwargs):
 try:
     from .recognizers import google, google_cloud
     from .recognizers.whisper_api import groq, openai
-    from .recognizers.whisper_local import whisper
+    from .recognizers.whisper_local import faster_whisper, whisper
 except (ModuleNotFoundError, ImportError):
     pass
 else:
     Recognizer.recognize_google = google.recognize_legacy
     Recognizer.recognize_google_cloud = google_cloud.recognize
     Recognizer.recognize_whisper = whisper.recognize
+    Recognizer.recognize_faster_whisper = faster_whisper.recognize
     Recognizer.recognize_openai = openai.recognize
     Recognizer.recognize_whisper_api = openai.recognize  # Deprecated
     Recognizer.recognize_groq = groq.recognize

From e23338aa26a99f1a34b8632d563cfa552350e539 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:32:39 +0900
Subject: [PATCH 16/22] [docs] faster-whisper installation

---
 README.rst | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/README.rst b/README.rst
index a05f0670..83a05a4d 100644
--- a/README.rst
+++ b/README.rst
@@ -97,6 +97,7 @@ To use all of the functionality of the library, you should have:
 * **FLAC encoder** (required only if the system is not x86-based Windows/Linux/OS X)
 * **Vosk** (required only if you need to use Vosk API speech recognition ``recognizer_instance.recognize_vosk``)
 * **Whisper** (required only if you need to use Whisper ``recognizer_instance.recognize_whisper``)
+* **Faster Whisper** (required only if you need to use Faster Whisper ``recognizer_instance.recognize_faster_whisper``)
 * **openai** (required only if you need to use OpenAI Whisper API speech recognition ``recognizer_instance.recognize_openai``)
 * **groq** (required only if you need to use Groq Whisper API speech recognition ``recognizer_instance.recognize_groq``)
 
@@ -179,6 +180,13 @@ Whisper is **required if and only if you want to use whisper** (``recognizer_ins
 
 You can install it with ``python3 -m pip install SpeechRecognition[whisper-local]``.
 
+Faster Whisper (for Faster Whisper users)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+The library `faster-whisper <https://pypi.org/project/faster-whisper/>`__ is **required if and only if you want to use Faster Whisper** (``recognizer_instance.recognize_faster_whisper``).
+
+You can install it with ``python3 -m pip install SpeechRecognition[faster-whisper]``.
+
 OpenAI Whisper API (for OpenAI Whisper API users) 
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

From 3fea94819113143e7856ca87e1a701ce4b9873bc Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 15:42:53 +0900
Subject: [PATCH 17/22] [bugfix] Install numpy in dev environment

* Fix Ubuntu 3.13 failed: ModuleNotFoundError: No module named 'numpy'
---
 setup.cfg | 1 +
 1 file changed, 1 insertion(+)

diff --git a/setup.cfg b/setup.cfg
index f3848d4d..9a4a4ce8 100644
--- a/setup.cfg
+++ b/setup.cfg
@@ -3,6 +3,7 @@ dev =
     pytest
     pytest-randomly
     respx
+    numpy
 audio =
     PyAudio >= 0.2.11
 pocketsphinx =

From 070d02f98650b24e3b36b496e9b33f8891509f6c Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Mon, 30 Dec 2024 18:08:15 +0900
Subject: [PATCH 18/22] [bugfix] Run (=skip) tests without faster-whisper

---
 .../whisper_local/test_faster_whisper.py      | 30 +++++++++++++++----
 1 file changed, 24 insertions(+), 6 deletions(-)

diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 06a6b0e6..a2d2ffc8 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -1,15 +1,12 @@
+from __future__ import annotations
+
 import sys
 from collections.abc import Generator
+from typing import TYPE_CHECKING
 from unittest.mock import ANY, MagicMock, patch
 
 import numpy as np
 import pytest
-from faster_whisper.transcribe import (
-    Segment,
-    TranscriptionInfo,
-    TranscriptionOptions,
-    VadOptions,
-)
 
 from speech_recognition import Recognizer
 from speech_recognition.audio import AudioData
@@ -17,6 +14,9 @@
     recognize,
 )
 
+if TYPE_CHECKING:
+    from faster_whisper.transcribe import Segment, TranscriptionInfo
+
 
 @pytest.fixture
 def audio_data() -> AudioData:
@@ -27,6 +27,8 @@ def audio_data() -> AudioData:
 
 @pytest.fixture
 def segment() -> Segment:
+    from faster_whisper.transcribe import Segment
+
     mocked_segment = MagicMock(spec=Segment(*[None] * 11))
     mocked_segment.text = ""
     return mocked_segment
@@ -34,6 +36,8 @@ def segment() -> Segment:
 
 @pytest.fixture
 def transcription_info() -> TranscriptionInfo:
+    from faster_whisper.transcribe import TranscriptionInfo
+
     return MagicMock(spec=TranscriptionInfo(*[None] * 7))
 
 
@@ -56,6 +60,13 @@ class TestTranscribe:
     def test_default_parameters(
         self, is_available, WhisperModel, audio_data, soundfile_read
     ):
+        from faster_whisper.transcribe import (
+            Segment,
+            TranscriptionInfo,
+            TranscriptionOptions,
+            VadOptions,
+        )
+
         sf_read, audio_array = soundfile_read
 
         def segments():
@@ -126,6 +137,13 @@ def segments_generator():
     def test_show_dict(
         self, is_available, WhisperModel, audio_data, soundfile_read
     ):
+        from faster_whisper.transcribe import (
+            Segment,
+            TranscriptionInfo,
+            TranscriptionOptions,
+            VadOptions,
+        )
+
         sf_read, audio_array = soundfile_read
 
         def segments():

From 0edcc439fcfef609b33422ebf8ce53c7453b8aa8 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Tue, 31 Dec 2024 00:08:22 +0900
Subject: [PATCH 19/22] [feat] Use device="auto" (default)

---
 .../whisper_local/faster_whisper.py           |  5 +--
 .../whisper_local/test_faster_whisper.py      | 37 ++-----------------
 2 files changed, 5 insertions(+), 37 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index ffc858b3..e8a41664 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -69,12 +69,9 @@ def recognize(
 
     Other values are passed directly to whisper. See https://github.com/SYSTRAN/faster-whisper/blob/master/faster_whisper/transcribe.py for all options.
     """
-    import torch
     from faster_whisper import WhisperModel
 
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-
-    model = WhisperModel(model, device=device)
+    model = WhisperModel(model)
     whisper_recognizer = WhisperCompatibleRecognizer(
         TranscribableAdapter(model)
     )
diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index a2d2ffc8..05883250 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -56,9 +56,8 @@ def soundfile_read() -> Generator[tuple[MagicMock, np.ndarray], None, None]:
 )
 @patch("faster_whisper.WhisperModel")
 class TestTranscribe:
-    @patch("torch.cuda.is_available", return_value=False)
     def test_default_parameters(
-        self, is_available, WhisperModel, audio_data, soundfile_read
+        self, WhisperModel, audio_data, soundfile_read
     ):
         from faster_whisper.transcribe import (
             Segment,
@@ -101,8 +100,7 @@ def segments():
         actual = recognize(MagicMock(spec=Recognizer), audio_data)
 
         assert actual == " 1, 2, 3"
-        is_available.assert_called_once_with()
-        WhisperModel.assert_called_once_with("base", device="cpu")
+        WhisperModel.assert_called_once_with("base")
         audio_data.get_wav_data.assert_called_once_with(convert_rate=16_000)
         sf_read.assert_called_once_with(ANY)
         assert sf_read.call_args[0][0].read() == b"audio data"
@@ -111,32 +109,7 @@ def segments():
             audio_array.astype.return_value
         )
 
-    @patch("torch.cuda.is_available", return_value=True)
-    def test_gpu_available(
-        self,
-        is_available,
-        WhisperModel,
-        audio_data,
-        segment,
-        transcription_info,
-        soundfile_read,
-    ):
-        def segments_generator():
-            yield segment
-
-        WhisperModel.return_value.transcribe.return_value = (
-            segments_generator(),
-            transcription_info,
-        )
-
-        _ = recognize(MagicMock(spec=Recognizer), audio_data)
-
-        WhisperModel.assert_called_once_with("base", device="cuda")
-
-    @patch("torch.cuda.is_available", return_value=False)
-    def test_show_dict(
-        self, is_available, WhisperModel, audio_data, soundfile_read
-    ):
+    def test_show_dict(self, WhisperModel, audio_data, soundfile_read):
         from faster_whisper.transcribe import (
             Segment,
             TranscriptionInfo,
@@ -199,10 +172,8 @@ def segments():
         }
         assert actual == expected
 
-    @patch("torch.cuda.is_available", return_value=False)
     def test_pass_parameters(
         self,
-        is_available,
         WhisperModel,
         audio_data,
         segment,
@@ -230,7 +201,7 @@ def segments_generator():
             beam_size=5,
         )
 
-        WhisperModel.assert_called_once_with("small", device="cpu")
+        WhisperModel.assert_called_once_with("small")
         whisper_model.transcribe.assert_called_once_with(
             audio_array.astype.return_value,
             language="fr",

From ef69c4a382f58d1a859522786bd5c453a7036995 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Tue, 31 Dec 2024 00:15:58 +0900
Subject: [PATCH 20/22] [feat] Run manually as e2e

python -m speech_recognition.recognizers.whisper_local.faster_whisper
---
 .../recognizers/whisper_local/faster_whisper.py | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index e8a41664..a6d59ae6 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -78,3 +78,20 @@ def recognize(
     return whisper_recognizer.recognize(
         audio_data, show_dict=show_dict, **transcribe_options
     )
+
+
+if __name__ == "__main__":
+    import argparse
+
+    import speech_recognition as sr
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("audio_file")
+    args = parser.parse_args()
+
+    r = sr.Recognizer()
+    with sr.AudioFile(args.audio_file) as source:
+        audio_data = r.listen(source)
+
+    transcription = recognize(None, audio_data)
+    print(transcription)

From ac556c15d6323e3654673364478754e9d2a3f904 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Tue, 31 Dec 2024 00:27:03 +0900
Subject: [PATCH 21/22] [feat] Enable to specify initialization parameters

---
 .../whisper_local/faster_whisper.py           | 11 +++++++-
 .../whisper_local/test_faster_whisper.py      | 27 +++++++++++++++++--
 2 files changed, 35 insertions(+), 3 deletions(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index a6d59ae6..48a2e696 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -36,6 +36,14 @@ def transcribe(
         }
 
 
+class InitOptionalParameters(TypedDict, total=False):
+    # https://github.com/SYSTRAN/faster-whisper/blob/v1.1.0/faster_whisper/transcribe.py#L575
+    device: Literal["cpu", "gpu", "auto"]
+    compute_type: str
+    download_root: str
+    # TODO Add others
+
+
 class TranscribeOptionalParameters(TypedDict, total=False):
     # https://github.com/SYSTRAN/faster-whisper/blob/v1.1.0/faster_whisper/transcribe.py#L692
     language: str
@@ -49,6 +57,7 @@ def recognize(
     audio_data: AudioData,
     model: str = "base",
     show_dict: bool = False,
+    init_options: InitOptionalParameters | None = None,
     **transcribe_options: Unpack[TranscribeOptionalParameters],
 ) -> str:
     """Performs speech recognition on ``audio_data`` (an ``AudioData`` instance), using Whisper.
@@ -71,7 +80,7 @@ def recognize(
     """
     from faster_whisper import WhisperModel
 
-    model = WhisperModel(model)
+    model = WhisperModel(model, **init_options or {})
     whisper_recognizer = WhisperCompatibleRecognizer(
         TranscribableAdapter(model)
     )
diff --git a/tests/recognizers/whisper_local/test_faster_whisper.py b/tests/recognizers/whisper_local/test_faster_whisper.py
index 05883250..e2eb7632 100644
--- a/tests/recognizers/whisper_local/test_faster_whisper.py
+++ b/tests/recognizers/whisper_local/test_faster_whisper.py
@@ -117,8 +117,6 @@ def test_show_dict(self, WhisperModel, audio_data, soundfile_read):
             VadOptions,
         )
 
-        sf_read, audio_array = soundfile_read
-
         def segments():
             yield Segment(
                 id=1,
@@ -208,3 +206,28 @@ def segments_generator():
             task="translate",
             beam_size=5,
         )
+
+    def test_init_parameters(
+        self,
+        WhisperModel,
+        audio_data,
+        segment,
+        transcription_info,
+        soundfile_read,
+    ):
+        def segments_generator():
+            yield segment
+
+        whisper_model = WhisperModel.return_value
+        whisper_model.transcribe.return_value = (
+            segments_generator(),
+            transcription_info,
+        )
+
+        _ = recognize(
+            MagicMock(spec=Recognizer),
+            audio_data,
+            init_options={"compute_type": "int8"},
+        )
+
+        WhisperModel.assert_called_once_with("base", compute_type="int8")

From e663cb0609b76afc64ee90883d1cc72ab6a951f7 Mon Sep 17 00:00:00 2001
From: ftnext <takuyafjp+develop@gmail.com>
Date: Tue, 31 Dec 2024 00:41:37 +0900
Subject: [PATCH 22/22] [docs] Fix type hint

---
 speech_recognition/recognizers/whisper_local/faster_whisper.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/speech_recognition/recognizers/whisper_local/faster_whisper.py b/speech_recognition/recognizers/whisper_local/faster_whisper.py
index 48a2e696..e7ce10fb 100644
--- a/speech_recognition/recognizers/whisper_local/faster_whisper.py
+++ b/speech_recognition/recognizers/whisper_local/faster_whisper.py
@@ -59,7 +59,7 @@ def recognize(
     show_dict: bool = False,
     init_options: InitOptionalParameters | None = None,
     **transcribe_options: Unpack[TranscribeOptionalParameters],
-) -> str:
+) -> str | TranscribeOutput:
     """Performs speech recognition on ``audio_data`` (an ``AudioData`` instance), using Whisper.
 
     Pick ``model`` size (Same as Whisper).