mirror of
https://github.com/EDeev/api_processor.git
synced 2026-06-15 19:11:01 +03:00
51 lines
1.8 KiB
Python
51 lines
1.8 KiB
Python
import os, wave, vosk, ffmpeg
|
|
|
|
MODEL_PATH = r"models/vosk-model-small-ru-0.22"
|
|
FFMPEG_PATH = r"models/ffmpeg/bin/ffmpeg.exe"
|
|
|
|
def convert_audio_to_wav(input_file, output_file, FFMPEG_PATH):
|
|
try:
|
|
(
|
|
ffmpeg
|
|
.input(input_file)
|
|
.output(output_file, format='wav', acodec='pcm_s16le', ar='16000', ac=1,
|
|
af='acompressor,afftdn,dynaudnorm,aresample=16000') # 16kHz для Vosk
|
|
.global_args('-loglevel', 'quiet')
|
|
.run(cmd=FFMPEG_PATH, overwrite_output=True)
|
|
)
|
|
print(f"Конвертация завершена: {output_file}")
|
|
except ffmpeg.Error as e:
|
|
print("Ошибка при конвертации:", e.stderr.decode())
|
|
|
|
|
|
vosk.SetLogLevel(-1)
|
|
|
|
def recognize_speech(audio_path) -> str:
|
|
if not os.path.exists(MODEL_PATH):
|
|
print("Ошибка: Модель не найдена!")
|
|
return ""
|
|
|
|
model = vosk.Model(MODEL_PATH)
|
|
|
|
if audio_path.split('.')[-1] != "wav":
|
|
convert.convert_audio_to_wav(audio_path, "audio.wav", FFMPEG_PATH)
|
|
audio_path = "audio.wav"
|
|
else:
|
|
with wave.open(audio_path, "rb") as wf:
|
|
if wf.getnchannels() != 1 or wf.getsampwidth() != 2 or wf.getframerate() != 16000:
|
|
convert.convert_audio_to_wav(audio_path, "audio.wav", FFMPEG_PATH)
|
|
audio_path = "audio.wav"
|
|
|
|
|
|
with wave.open(audio_path, "rb") as wf: # использование vosk
|
|
recognizer = vosk.KaldiRecognizer(model, wf.getframerate())
|
|
while True:
|
|
data = wf.readframes(3200)
|
|
if not data:
|
|
break
|
|
recognizer.AcceptWaveform(data)
|
|
|
|
if audio_path == "audio.wav":
|
|
os.remove(audio_path)
|
|
|
|
return recognizer.FinalResult().split(": \"")[-1][:-3]
|