#187 Set condition_on_previous_text for whisper modules

This commit is contained in:
Cohee
2023-11-19 19:19:45 +02:00
parent 289bb2e86b
commit 61efbb9aa3
3 changed files with 12 additions and 11 deletions

View File

@@ -45,7 +45,7 @@ def load_model(file_path=None):
return (whisper.load_model("base.en"), vosk.Model(lang="en-us"))
else:
return (whisper.load_model(file_path), vosk.Model(lang="en-us"))
def convert_bytearray_to_wav_ndarray(input_bytearray: bytes, sampling_rate=16000):
"""
Convert a bytearray to wav format to output in a file for quality check debuging
@@ -65,7 +65,7 @@ def record_and_transcript():
if whisper_model is None:
print(DEBUG_PREFIX,"Whisper model not initialized yet.")
return ""
q = queue.Queue()
stream_errors = list()
@@ -91,7 +91,7 @@ def record_and_transcript():
data = q.get()
if len(stream_errors) > 0:
raise Exception(DEBUG_PREFIX+" Stream errors: "+str(stream_errors))
full_recording.extend(data)
if rec.AcceptWaveform(data):
@@ -105,9 +105,9 @@ def record_and_transcript():
output_file = convert_bytearray_to_wav_ndarray(input_bytearray=full_recording, sampling_rate=samplerate)
sf.write(file=RECORDING_FILE_PATH, data=output_file, samplerate=samplerate)
print(DEBUG_PREFIX, "Recorded message saved to", RECORDING_FILE_PATH)
# Whisper HACK
result = whisper_model.transcribe(RECORDING_FILE_PATH)
result = whisper_model.transcribe(RECORDING_FILE_PATH, condition_on_previous_text=False)
transcript = result["text"]
print(DEBUG_PREFIX, "Transcripted from audio file (whisper):", transcript)
# ----------------------------------
@@ -118,4 +118,4 @@ def record_and_transcript():
except Exception as e: # No exception observed during test but we never know
print(e)
abort(500, DEBUG_PREFIX+" Exception occurs while recording")
abort(500, DEBUG_PREFIX+" Exception occurs while recording")