flare / stt /stt_google.py
ciyidogan's picture
Update stt/stt_google.py
c5bf788 verified
raw
history blame
20.8 kB
"""
Google Cloud Speech-to-Text Implementation
"""
import os
import asyncio
from typing import AsyncIterator, AsyncGenerator, Optional, List, Any
import numpy as np
from datetime import datetime
import sys
import queue
import threading
import time
import traceback
from utils.logger import log_info, log_error, log_debug, log_warning
# Import Google Cloud Speech only if available
try:
from google.cloud import speech
from google.api_core import exceptions
GOOGLE_SPEECH_AVAILABLE = True
except ImportError:
GOOGLE_SPEECH_AVAILABLE = False
log_info("⚠️ Google Cloud Speech library not installed")
from .stt_interface import STTInterface, STTConfig, TranscriptionResult
class GoogleCloudSTT(STTInterface):
"""Google Cloud Speech-to-Text implementation"""
def __init__(self, credentials_path: Optional[str] = None):
"""Initialize Google Cloud STT"""
log_info("🎀 Creating STT provider: google")
# Initialize all required attributes
self.client = None
self.streaming_config = None
self.stream_thread = None
self.audio_queue = queue.Queue()
self.responses_queue = queue.Queue()
self.is_streaming = False
self.should_stop = False
self.error_message = None
self.session_id = 0
self.stream_start_time = None
# Additional attributes
self.lock = threading.Lock()
self.single_utterance = False
self.chunk_count = 0
self.total_bytes = 0
self.stop_event = threading.Event()
# Set Google credentials
if credentials_path:
if os.path.exists(credentials_path):
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = credentials_path
log_info(f"βœ… Google credentials set from: {credentials_path}")
else:
log_error(f"❌ Credentials file not found: {credentials_path}")
raise ValueError(f"Google credentials file not found: {credentials_path}")
else:
# Fallback to environment variable
creds_path = os.environ.get("GOOGLE_APPLICATION_CREDENTIALS")
if not creds_path:
creds_path = "./credentials/google-service-account.json"
if os.path.exists(creds_path):
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = creds_path
log_info(f"βœ… Google credentials set from default: {creds_path}")
else:
raise ValueError("Google credentials not found. Please provide credentials_path")
# Test credentials
try:
log_info("πŸ” Testing Google credentials...")
test_client = speech.SpeechClient()
log_info("βœ… Google credentials valid")
except Exception as e:
log_error(f"❌ Invalid Google credentials: {e}")
raise
def _get_encoding(self, encoding_str: str):
"""Convert encoding string to Google Speech enum"""
if not GOOGLE_SPEECH_AVAILABLE:
return None
encoding_map = {
"WEBM_OPUS": speech.RecognitionConfig.AudioEncoding.WEBM_OPUS,
"LINEAR16": speech.RecognitionConfig.AudioEncoding.LINEAR16,
"FLAC": speech.RecognitionConfig.AudioEncoding.FLAC,
"MP3": speech.RecognitionConfig.AudioEncoding.MP3,
"OGG_OPUS": speech.RecognitionConfig.AudioEncoding.OGG_OPUS,
}
return encoding_map.get(encoding_str, speech.RecognitionConfig.AudioEncoding.WEBM_OPUS)
# Alias for compatibility
_get_google_encoding = _get_encoding
async def stream_audio(self, audio_chunk: bytes) -> AsyncGenerator[TranscriptionResult, None]:
"""Stream audio chunk and get results"""
if not self.is_streaming:
raise Exception("Streaming not started")
try:
chunk_size = len(audio_chunk)
# Log first chunk details
if self.chunk_count == 0:
log_info(f"πŸ“€ First chunk - size: {chunk_size} bytes")
if audio_chunk.startswith(b'\x1a\x45\xdf\xa3'):
log_info("βœ… Valid WEBM header detected")
else:
hex_preview = audio_chunk[:20].hex()
log_warning(f"⚠️ Unexpected audio format. First 20 bytes: {hex_preview}")
# Try to measure audio level (if it's raw PCM)
try:
if encoding_str == "LINEAR16": # Only for raw PCM
audio_array = np.frombuffer(audio_chunk, dtype=np.int16)
if len(audio_array) > 0:
rms = np.sqrt(np.mean(audio_array.astype(float) ** 2))
db = 20 * np.log10(max(rms, 1) / 32768.0)
if self.chunk_count % 50 == 0:
log_info(f"πŸ”Š Audio level: {db:.1f} dB")
except:
pass
# Put chunk in queue
self.audio_queue.put(audio_chunk)
self.chunk_count += 1
self.total_bytes += chunk_size
# Log progress
if self.chunk_count % 50 == 0:
log_info(f"πŸ“€ Progress: {self.chunk_count} chunks, {self.total_bytes/1024:.1f}KB total")
# Check for responses
timeout = 0.1
end_time = time.time() + timeout
while time.time() < end_time:
try:
result = self.responses_queue.get_nowait()
log_info(f"🎯 Got result from queue: is_final={result.is_final}, text='{result.text[:30]}...'")
yield result
except queue.Empty:
await asyncio.sleep(0.01)
except Exception as e:
log_error(f"Error getting result from queue: {e}")
break
except Exception as e:
log_error(f"❌ Error in stream_audio: {e}")
raise
async def stop_streaming(self) -> Optional[TranscriptionResult]:
"""Stop streaming and clean up all resources"""
if not self.is_streaming and not self.stream_thread:
log_debug("Already stopped, nothing to do")
return None
try:
log_info(f"πŸ›‘ Stopping Google STT streaming session #{self.session_id}")
# Set flags
self.is_streaming = False
self.should_stop = True
self.stop_event.set()
# Send poison pill
if self.audio_queue:
try:
self.audio_queue.put(None)
except:
pass
# Wait for thread
if self.stream_thread and self.stream_thread.is_alive():
log_info("⏳ Waiting for stream thread to finish...")
self.stream_thread.join(timeout=5.0)
if self.stream_thread.is_alive():
log_warning("⚠️ STT thread did not stop gracefully after 5s")
else:
log_info("βœ… Stream thread finished")
# Get final result
final_result = None
if self.responses_queue:
while not self.responses_queue.empty():
try:
result = self.responses_queue.get_nowait()
if result.is_final:
final_result = result
except queue.Empty:
break
# Close client
if self.client:
try:
if hasattr(self.client, 'transport') and hasattr(self.client.transport, 'close'):
self.client.transport.close()
log_debug("βœ… Client transport closed")
if hasattr(self.client, '_transport') and hasattr(self.client._transport, '_grpc_channel'):
self.client._transport._grpc_channel.close()
log_debug("βœ… gRPC channel closed")
except Exception as e:
log_warning(f"⚠️ Error closing Google client: {e}")
finally:
self.client = None
# Reset state
self.audio_queue = None
self.responses_queue = None
self.stream_thread = None
self.streaming_config = None
self.stop_event.clear()
log_info(f"βœ… Google STT streaming session #{self.session_id} stopped and cleaned")
return final_result
except Exception as e:
log_error(f"❌ Error during stop_streaming", error=str(e))
self.is_streaming = False
self.stream_thread = None
self.client = None
self.streaming_config = None
self.stop_event.clear()
self.audio_queue = None
self.responses_queue = None
return None
def supports_realtime(self) -> bool:
"""Google Cloud STT supports real-time streaming"""
return True
def get_supported_languages(self) -> List[str]:
"""Get list of supported language codes"""
return [
"tr-TR", "en-US", "en-GB", "de-DE", "fr-FR", "es-ES",
"it-IT", "pt-BR", "ru-RU", "ja-JP", "ko-KR", "zh-CN", "ar-SA"
]
def get_provider_name(self) -> str:
"""Get provider name"""
return "google"
def _reset_session(self):
"""Reset session data"""
# Clear queues
while not self.audio_queue.empty():
try:
self.audio_queue.get_nowait()
except queue.Empty:
break
while not self.responses_queue.empty():
try:
self.responses_queue.get_nowait()
except queue.Empty:
break
# Reset state
self.should_stop = False
self.error_message = None
self.session_id += 1
self.stream_start_time = time.time()
self.chunk_count = 0
self.total_bytes = 0
log_info(f"πŸ”„ Google STT session data reset. New session ID: {self.session_id}")
# Create fresh queues
self.audio_queue = queue.Queue()
self.responses_queue = queue.Queue()
log_debug("βœ… Created fresh queues")
def _create_fresh_queues(self):
"""Create fresh queue instances"""
if self.audio_queue:
while not self.audio_queue.empty():
try:
self.audio_queue.get_nowait()
except:
pass
if self.responses_queue:
while not self.responses_queue.empty():
try:
self.responses_queue.get_nowait()
except:
pass
self.audio_queue = queue.Queue(maxsize=1000)
self.responses_queue = queue.Queue(maxsize=100)
log_debug("βœ… Created fresh queues")
def _request_generator(self):
"""Generate requests for the streaming recognize API"""
# First request with config
yield speech.StreamingRecognizeRequest(streaming_config=self.streaming_config)
# Audio chunks
while not self.should_stop:
try:
audio_chunk = self.audio_queue.get(timeout=0.1)
if audio_chunk is None:
log_info("πŸ“› Poison pill received, stopping request generator")
break
yield speech.StreamingRecognizeRequest(audio_content=audio_chunk)
except queue.Empty:
continue
except Exception as e:
log_error(f"Error in request generator: {e}")
break
log_info(f"πŸ“Š Request generator finished. Total chunks: {self.chunk_count}, Total bytes: {self.total_bytes}")
async def start_streaming(self, config: STTConfig) -> None:
"""Initialize streaming session with clean state"""
try:
# Thread safety iΓ§in lock kullan
async with asyncio.Lock():
# Clean up any existing stream
if self.is_streaming or self.stream_thread:
log_warning("⚠️ Previous stream still active, stopping it first")
await self.stop_streaming()
await asyncio.sleep(0.5)
# Double-check after cleanup
if self.stream_thread and self.stream_thread.is_alive():
log_error(f"❌ Stream thread STILL running after cleanup! Thread: {self.stream_thread.name}")
raise Exception("Failed to stop previous stream thread")
# Reset session
self._reset_session()
self.single_utterance = config.single_utterance
self.current_encoding = config.encoding
log_info(f"🎀 Starting Google STT streaming session #{self.session_id} with config: {config}")
# Create fresh queues
self._create_fresh_queues()
self.stop_event.clear()
self.should_stop = False
# Create new client
self.client = speech.SpeechClient()
log_info("βœ… Created new Google Speech client")
# Create recognition config
recognition_config = speech.RecognitionConfig(
encoding=speech.RecognitionConfig.AudioEncoding.WEBM_OPUS,
sample_rate_hertz=16000,
language_code="tr-TR",
enable_automatic_punctuation=True,
model="latest_long",
use_enhanced=True,
max_alternatives=1,
metadata=speech.RecognitionMetadata(
interaction_type=speech.RecognitionMetadata.InteractionType.VOICE_SEARCH,
microphone_distance=speech.RecognitionMetadata.MicrophoneDistance.NEARFIELD,
recording_device_type=speech.RecognitionMetadata.RecordingDeviceType.PC,
)
)
# Create streaming config with VAD
self.streaming_config = speech.StreamingRecognitionConfig(
config=recognition_config,
interim_results=True,
single_utterance=False,
enable_voice_activity_events=True # βœ… VAD events enabled
)
self.is_streaming = True
self.stop_event.clear()
# Thread başlatmadan ânce son kontrol
if self.stream_thread is not None:
log_error("❌ stream_thread should be None at this point!")
self.stream_thread = None
self.is_streaming = True
# Start streaming thread with unique ID
thread_id = f"GoogleSTT-Session-{self.session_id}-{int(time.time()*1000)}"
self.stream_thread = threading.Thread(
target=self._run_stream,
name=thread_id
)
self.stream_thread.daemon = True
log_info(f"πŸš€ Starting thread: {thread_id}")
self.stream_thread.start()
log_info(f"βœ… Google STT streaming session #{self.session_id} started successfully")
except Exception as e:
log_error(f"❌ Failed to start Google STT streaming", error=str(e))
self.is_streaming = False
self.client = None
self._create_fresh_queues()
raise
def _run_stream(self):
"""Run the streaming recognition loop in a separate thread"""
try:
thread_id = threading.current_thread().ident
log_info(f"🎀 Google STT stream thread started - Thread ID: {thread_id}, Session: {self.session_id}")
# Create request generator
requests = self._request_generator()
# Create streaming client
log_info(f"🎀 Creating Google STT streaming client... Thread ID: {thread_id}")
# Get responses (no timeout parameter!)
responses = self.client.streaming_recognize(requests)
# Track responses
first_response_time = None
response_count = 0
# Process responses
for response in responses:
if self.should_stop:
log_info("πŸ›‘ Stop flag detected, ending stream")
break
response_count += 1
if first_response_time is None:
first_response_time = time.time()
elapsed = first_response_time - self.stream_start_time
log_info(f"πŸŽ‰ FIRST RESPONSE from Google STT after {elapsed:.2f}s")
# Check for VAD events
if hasattr(response, 'speech_event_type') and response.speech_event_type:
event_type = response.speech_event_type
log_info(f"πŸŽ™οΈ VAD Event: {event_type}")
if event_type == speech.StreamingRecognizeResponse.SpeechEventType.END_OF_SINGLE_UTTERANCE:
log_info("πŸ”š End of utterance detected by VAD")
# Log response
has_results = len(response.results) > 0 if hasattr(response, 'results') else False
log_info(f"πŸ“¨ Google STT Response #{response_count}: has_results={has_results}")
if not response.results:
continue
# Process results
for result_idx, result in enumerate(response.results):
# Check result type
result_type = "πŸ”„ INTERIM" if not result.is_final else "βœ… FINAL"
stability = getattr(result, 'stability', 0.0)
log_info(f"{result_type} Result #{result_idx}: "
f"alternatives={len(result.alternatives)}, "
f"stability={stability:.3f}")
if result.alternatives:
best_alternative = result.alternatives[0]
transcript = best_alternative.transcript
confidence = best_alternative.confidence if result.is_final else stability
# Log transcript
if result.is_final:
log_info(f"βœ… FINAL TRANSCRIPT: '{transcript}' "
f"(confidence: {confidence:.3f})")
else:
log_info(f"πŸ”„ INTERIM TRANSCRIPT: '{transcript[:100]}...' "
f"(stability: {stability:.3f})")
# Queue result
result_obj = TranscriptionResult(
text=transcript,
is_final=result.is_final,
confidence=confidence,
timestamp=datetime.utcnow()
)
self.responses_queue.put(result_obj)
log_info(f"πŸ“₯ {'FINAL' if result.is_final else 'INTERIM'} result queued")
# Log completion
if response_count == 0:
log_error("❌ Google STT stream ended without ANY responses!")
else:
log_info(f"βœ… Google STT stream ended normally after {response_count} responses")
except Exception as e:
log_error(f"❌ Google STT error: {e}")
if hasattr(e, 'details'):
log_error(f"Error details: {e.details}")
self.error_message = str(e)
finally:
log_info("🎀 Google STT stream thread ended")
with self.lock:
self.is_streaming = False