every_holiday/VRCT
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

🚧[WIP/TEST] Model : 文字起こし起動時にエンジンを選択するように変更#2

Browse Source
This commit is contained in:
misyaguziya
2024-02-13 23:25:31 +09:00
parent 07b3c92f1b
commit 7cdd9d19d7
2 changed files with 6 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
model.py
View File

@@ -346,7 +346,7 @@ class Model:
whisper_weight_type=config.WHISPER_WEIGHT_TYPE, whisper_weight_type=config.WHISPER_WEIGHT_TYPE,
) )
def sendMicTranscript(): def sendMicTranscript():
mic_transcriber.transcribeAudioQueue(mic_audio_queue, config.SOURCE_LANGUAGE, config.SOURCE_COUNTRY, config.SELECTED_TRANSCRIPTION_ENGINE) mic_transcriber.transcribeAudioQueue(mic_audio_queue, config.SOURCE_LANGUAGE, config.SOURCE_COUNTRY)
message = mic_transcriber.getTranscript() message = mic_transcriber.getTranscript()
try: try:
fnc(message) fnc(message)
@@ -449,7 +449,7 @@ class Model:
whisper_weight_type=config.WHISPER_WEIGHT_TYPE, whisper_weight_type=config.WHISPER_WEIGHT_TYPE,
) )
def sendSpeakerTranscript(): def sendSpeakerTranscript():
speaker_transcriber.transcribeAudioQueue(speaker_audio_queue, config.TARGET_LANGUAGE, config.TARGET_COUNTRY, config.SELECTED_TRANSCRIPTION_ENGINE) speaker_transcriber.transcribeAudioQueue(speaker_audio_queue, config.TARGET_LANGUAGE, config.TARGET_COUNTRY)
message = speaker_transcriber.getTranscript() message = speaker_transcriber.getTranscript()
try: try:
fnc(message) fnc(message)

13
models/transcription/transcription_transcriber.py
View File

@@ -37,21 +37,16 @@ class AudioTranscriber:
self.whisper_model = getWhisperModel(root, whisper_weight_type) self.whisper_model = getWhisperModel(root, whisper_weight_type)
self.transcription_engine = "Whisper" self.transcription_engine = "Whisper"
def transcribeAudioQueue(self, audio_queue, language, country, transcription_engine): def transcribeAudioQueue(self, audio_queue, language, country):
audio, time_spoken = audio_queue.get() audio, time_spoken = audio_queue.get()
self.updateLastSampleAndPhraseStatus(audio, time_spoken) self.updateLastSampleAndPhraseStatus(audio, time_spoken)
text = '' text = ''
try: try:
# Whisperが使用できない場合はGoogle Speech-to-Textを使用する
if transcription_engine == "Whisper":
if self.whisper_model is None:
transcription_engine = "Google"
audio_data = self.audio_sources["process_data_func"]() audio_data = self.audio_sources["process_data_func"]()
match transcription_engine: match self.transcription_engine:
case "Google": case "Google":
text = self.audio_recognizer.recognize_google(audio_data, language=transcription_lang[language][country][transcription_engine]) text = self.audio_recognizer.recognize_google(audio_data, language=transcription_lang[language][country][self.transcription_engine])
case "Whisper": case "Whisper":
audio_data = np.frombuffer(audio_data.get_raw_data(convert_rate=16000, convert_width=2), np.int16).flatten().astype(np.float32) / 32768.0 audio_data = np.frombuffer(audio_data.get_raw_data(convert_rate=16000, convert_width=2), np.int16).flatten().astype(np.float32) / 32768.0
if isinstance(audio_data, torch.Tensor): if isinstance(audio_data, torch.Tensor):
@@ -62,7 +57,7 @@ class AudioTranscriber:
temperature=0.0, temperature=0.0,
log_prob_threshold=-0.8, log_prob_threshold=-0.8,
no_speech_threshold=0.6, no_speech_threshold=0.6,
language=transcription_lang[language][country][transcription_engine], language=transcription_lang[language][country][self.transcription_engine],
word_timestamps=False, word_timestamps=False,
without_timestamps=True, without_timestamps=True,
task="transcribe", task="transcribe",