mirror of
https://github.com/hackingthemarkets/openai-whisper-voice-commands.git
synced 2023-04-08 21:42:32 +03:00
38 lines
1.2 KiB
Python
38 lines
1.2 KiB
Python
import config
|
|
import whisper
|
|
import os, glob
|
|
|
|
# find most recent files in a directory
|
|
recordings_dir = os.path.join('recordings', '*')
|
|
|
|
model = whisper.load_model("base")
|
|
|
|
# list to store which wav files have been transcribed
|
|
transcribed = []
|
|
|
|
while True:
|
|
# get most recent wav recording in the recordings directory
|
|
files = sorted(glob.iglob(recordings_dir), key=os.path.getctime, reverse=True)
|
|
if len(files) < 1:
|
|
continue
|
|
|
|
latest_recording = files[0]
|
|
latest_recording_filename = latest_recording.split('/')[1]
|
|
|
|
if os.path.exists(latest_recording) and not latest_recording in transcribed:
|
|
audio = whisper.load_audio(latest_recording)
|
|
audio = whisper.pad_or_trim(audio)
|
|
mel = whisper.log_mel_spectrogram(audio).to(model.device)
|
|
options = whisper.DecodingOptions(language= 'en', fp16=False)
|
|
|
|
result = whisper.decode(model, mel, options)
|
|
|
|
if result.no_speech_prob < 0.5:
|
|
print(result.text)
|
|
|
|
# append text to transcript file
|
|
with open(config.TRANSCRIPT_FILE, 'a') as f:
|
|
f.write(result.text)
|
|
|
|
# save list of transcribed recordings so that we don't transcribe the same one again
|
|
transcribed.append(latest_recording) |