Spaces:
Building
Building
File size: 8,242 Bytes
0b9eed5 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 |
"""
Base STT Implementation
======================
Common audio processing and validation for all STT providers
"""
import struct
from typing import Optional, Tuple, List
from datetime import datetime
from abc import ABC, abstractmethod
from .stt_interface import STTInterface, STTConfig, TranscriptionResult
from utils.logger import log_info, log_error, log_debug, log_warning
class STTBase(STTInterface, ABC):
"""Base class for all STT implementations with common audio processing"""
def __init__(self):
super().__init__()
async def transcribe(self, audio_data: bytes, config: STTConfig) -> Optional[TranscriptionResult]:
"""Main transcription method with preprocessing"""
try:
# 1. Validate input
if not audio_data:
log_warning("⚠️ No audio data provided")
return None
log_info(f"📊 Transcribing {len(audio_data)} bytes of audio")
# 2. Analyze and validate audio
analysis_result = self._analyze_audio(audio_data, config.sample_rate)
if not analysis_result.is_valid:
log_warning(f"⚠️ Audio validation failed: {analysis_result.reason}")
return None
# 3. Preprocess audio if needed
processed_audio = self._preprocess_audio(audio_data, config)
# 4. Call provider-specific implementation
return await self._transcribe_impl(processed_audio, config, analysis_result)
except Exception as e:
log_error(f"❌ Error during transcription: {str(e)}")
import traceback
log_error(f"Traceback: {traceback.format_exc()}")
return None
@abstractmethod
async def _transcribe_impl(self, audio_data: bytes, config: STTConfig, analysis: 'AudioAnalysis') -> Optional[TranscriptionResult]:
"""Provider-specific transcription implementation"""
pass
def _analyze_audio(self, audio_data: bytes, sample_rate: int) -> 'AudioAnalysis':
"""Analyze audio quality and content"""
try:
samples = struct.unpack(f'{len(audio_data)//2}h', audio_data)
total_samples = len(samples)
# Basic statistics
non_zero_samples = [s for s in samples if s != 0]
zero_count = total_samples - len(non_zero_samples)
if non_zero_samples:
avg_amplitude = sum(abs(s) for s in non_zero_samples) / len(non_zero_samples)
max_amplitude = max(abs(s) for s in non_zero_samples)
else:
avg_amplitude = 0
max_amplitude = 0
log_info(f"🔍 Audio stats: {total_samples} total samples, {zero_count} zeros ({zero_count/total_samples:.1%})")
log_info(f"🔍 Non-zero stats: avg={avg_amplitude:.1f}, max={max_amplitude}")
# Section analysis (10 sections)
section_size = total_samples // 10
sections = []
for i in range(10):
start_idx = i * section_size
end_idx = (i + 1) * section_size if i < 9 else total_samples
section = samples[start_idx:end_idx]
section_non_zero = [s for s in section if s != 0]
section_max = max(abs(s) for s in section_non_zero) if section_non_zero else 0
section_avg = sum(abs(s) for s in section_non_zero) / len(section_non_zero) if section_non_zero else 0
zero_ratio = (len(section) - len(section_non_zero)) / len(section)
sections.append({
'max': section_max,
'avg': section_avg,
'zero_ratio': zero_ratio
})
log_info(f" Section {i+1}: max={section_max}, avg={section_avg:.1f}, zeros={zero_ratio:.1%}")
# Find speech start
speech_start_idx = self._find_speech_start(samples, sample_rate)
speech_start_time = speech_start_idx / sample_rate if speech_start_idx >= 0 else -1
if speech_start_idx >= 0:
log_info(f"🎤 Speech detected starting at sample {speech_start_idx} ({speech_start_time:.2f}s)")
else:
log_warning("⚠️ No speech detected above threshold in entire audio")
# Validation
is_valid = True
reason = ""
if max_amplitude < 100:
is_valid = False
reason = f"Audio appears silent: max_amplitude={max_amplitude}"
elif zero_count / total_samples > 0.95:
is_valid = False
reason = f"Audio is mostly zeros: {zero_count/total_samples:.1%}"
elif speech_start_idx < 0:
is_valid = False
reason = "No speech detected"
return AudioAnalysis(
total_samples=total_samples,
sample_rate=sample_rate,
zero_count=zero_count,
avg_amplitude=avg_amplitude,
max_amplitude=max_amplitude,
sections=sections,
speech_start_idx=speech_start_idx,
speech_start_time=speech_start_time,
is_valid=is_valid,
reason=reason
)
except Exception as e:
log_error(f"Audio analysis failed: {e}")
return AudioAnalysis(
total_samples=0,
sample_rate=sample_rate,
is_valid=False,
reason=f"Analysis failed: {e}"
)
def _find_speech_start(self, samples: List[int], sample_rate: int, threshold: int = 500) -> int:
"""Find the starting point of speech in audio"""
window_size = 100
for i in range(0, len(samples) - window_size, window_size):
window = samples[i:i + window_size]
rms = (sum(s * s for s in window) / window_size) ** 0.5
if rms > threshold:
return i
return -1
def _preprocess_audio(self, audio_data: bytes, config: STTConfig) -> bytes:
"""Preprocess audio if needed (can be overridden by providers)"""
# Default: no preprocessing
return audio_data
def _clean_audio_silence(self, audio_data: bytes, threshold: int = 50) -> bytes:
"""Remove leading/trailing silence"""
try:
samples = struct.unpack(f'{len(audio_data)//2}h', audio_data)
# Find first non-silent sample
start_idx = 0
for i, sample in enumerate(samples):
if abs(sample) > threshold:
start_idx = i
break
# Find last non-silent sample
end_idx = len(samples) - 1
for i in range(len(samples) - 1, -1, -1):
if abs(samples[i]) > threshold:
end_idx = i
break
# Add padding
start_idx = max(0, start_idx - 100)
end_idx = min(len(samples) - 1, end_idx + 100)
# Convert back
cleaned_samples = samples[start_idx:end_idx + 1]
cleaned_audio = struct.pack(f'{len(cleaned_samples)}h', *cleaned_samples)
log_debug(f"Audio cleaning: {len(audio_data)} → {len(cleaned_audio)} bytes")
return cleaned_audio
except Exception as e:
log_warning(f"Audio cleaning failed: {e}, using original")
return audio_data
@dataclass
class AudioAnalysis:
"""Audio analysis results"""
total_samples: int = 0
sample_rate: int = 16000
zero_count: int = 0
avg_amplitude: float = 0.0
max_amplitude: int = 0
sections: List[dict] = field(default_factory=list)
speech_start_idx: int = -1
speech_start_time: float = -1.0
is_valid: bool = False
reason: str = "" |