Ubuntu
143 lines
4.6 KiB
Python
143 lines
4.6 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
语音转文字 - FunASR 版本（阿里达摩院）
|
|
支持中文识别、自动标点
|
|
"""
|
|
import os
|
|
import subprocess
|
|
import warnings
|
|
|
|
warnings.filterwarnings('ignore')
|
|
|
|
UPLOAD_FOLDER = '/home/ubuntu/nanocode/audio-recorder/uploads'
|
|
MODEL_CACHE = os.path.expanduser('~/.cache/modelscope')
|
|
|
|
class FunASRRecognizer:
|
|
def __init__(self):
|
|
self.model = None
|
|
self.model_type = 'funasr' if self._check_model_cached() else 'vosk'
|
|
|
|
def _check_model_cached(self):
|
|
"""检查 FunASR 模型是否已下载"""
|
|
# 检查常用的模型路径
|
|
possible_paths = [
|
|
os.path.join(MODEL_CACHE, 'hub', 'damo', 'speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'),
|
|
os.path.join(MODEL_CACHE, 'hub', 'iic', 'speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch'),
|
|
]
|
|
return any(os.path.exists(p) for p in possible_paths)
|
|
|
|
def _load_model(self):
|
|
"""加载 FunASR 模型"""
|
|
if self.model is None:
|
|
try:
|
|
from funasr import AutoModel
|
|
print("🔄 加载 FunASR 模型...")
|
|
self.model = AutoModel(
|
|
model="paraformer-zh",
|
|
vad_model="fsmn-vad",
|
|
punc_model="ct-punc",
|
|
)
|
|
print("✅ FunASR 模型加载完成")
|
|
self.model_type = 'funasr'
|
|
except Exception as e:
|
|
print(f"⚠️ FunASR 加载失败: {e}, 回退到 Vosk")
|
|
self.model_type = 'vosk'
|
|
raise
|
|
return self.model
|
|
|
|
def convert_to_wav(self, input_path):
|
|
"""转换为 WAV 格式"""
|
|
output_path = input_path + '.wav'
|
|
try:
|
|
cmd = [
|
|
'ffmpeg', '-y', '-i', input_path,
|
|
'-ar', '16000', '-ac', '1', '-f', 'wav',
|
|
output_path
|
|
]
|
|
result = subprocess.run(cmd, capture_output=True, timeout=30)
|
|
if result.returncode == 0 and os.path.exists(output_path):
|
|
return output_path
|
|
except Exception as e:
|
|
print(f"转换错误: {e}")
|
|
return None
|
|
|
|
def transcribe(self, audio_path):
|
|
"""转写音频"""
|
|
# 先转换为 wav
|
|
wav_path = self.convert_to_wav(audio_path)
|
|
if not wav_path:
|
|
return None, "音频转换失败"
|
|
|
|
try:
|
|
# 尝试使用 FunASR
|
|
model = self._load_model()
|
|
|
|
if self.model_type == 'funasr':
|
|
# FunASR 识别
|
|
res = model.generate(input=wav_path, batch_size_s=300)
|
|
text = res[0]["text"]
|
|
else:
|
|
# 回退到 Vosk
|
|
text, error = self._transcribe_vosk(wav_path)
|
|
if error:
|
|
raise Exception(error)
|
|
|
|
# 清理临时文件
|
|
if os.path.exists(wav_path):
|
|
os.remove(wav_path)
|
|
|
|
return text, None
|
|
|
|
except Exception as e:
|
|
# 清理临时文件
|
|
if os.path.exists(wav_path):
|
|
os.remove(wav_path)
|
|
return None, str(e)
|
|
|
|
def _transcribe_vosk(self, wav_path):
|
|
"""备用：Vosk 识别"""
|
|
try:
|
|
from vosk import Model, KaldiRecognizer
|
|
import json
|
|
import wave
|
|
|
|
model_path = '/home/ubuntu/nanocode/audio-recorder/models/cn-model'
|
|
model = Model(model_path)
|
|
|
|
wf = wave.open(wav_path, 'rb')
|
|
rec = KaldiRecognizer(model, wf.getframerate())
|
|
|
|
results = []
|
|
while True:
|
|
data = wf.readframes(4000)
|
|
if len(data) == 0:
|
|
break
|
|
if rec.AcceptWaveform(data):
|
|
result = json.loads(rec.Result())
|
|
if result.get('text'):
|
|
results.append(result['text'])
|
|
|
|
final = json.loads(rec.FinalResult())
|
|
if final.get('text'):
|
|
results.append(final['text'])
|
|
|
|
wf.close()
|
|
return ' '.join(results), None
|
|
|
|
except Exception as e:
|
|
return None, f"Vosk 错误: {str(e)}"
|
|
|
|
# 全局识别器
|
|
_recognizer = None
|
|
|
|
def get_recognizer():
|
|
global _recognizer
|
|
if _recognizer is None:
|
|
_recognizer = FunASRRecognizer()
|
|
return _recognizer
|
|
|
|
def transcribe_audio(audio_path):
|
|
"""转写音频入口"""
|
|
recognizer = get_recognizer()
|
|
return recognizer.transcribe(audio_path)
|