let5see/AI-Video
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

优化edgetts的重试逻辑

Browse Source
This commit is contained in:
puke
2025-10-28 14:16:12 +08:00
committed by puke
parent 18b4152f39
commit 3d8cbe72e2
1 changed files with 115 additions and 64 deletions
Download Patch File Download Diff File Expand all files Collapse all files

179
reelforge/utils/tts_util.py
View File

@@ -7,6 +7,7 @@ Currently, TTS service uses ComfyUI workflows only.
import asyncio
import ssl
import random
import edge_tts as edge_tts_sdk
from loguru import logger
from aiohttp import WSServerHandshakeError, ClientResponseError
@@ -16,8 +17,16 @@ from aiohttp import WSServerHandshakeError, ClientResponseError
_SSL_VERIFY_ENABLED = False
# Retry configuration for Edge TTS (to handle 401 errors)
_RETRY_COUNT = 3 # Default retry count
_RETRY_DELAY = 2.0 # Retry delay in seconds
_RETRY_COUNT = 5 # Default retry count (increased from 3 to 5)
_RETRY_BASE_DELAY = 1.0 # Base retry delay in seconds (for exponential backoff)
_MAX_RETRY_DELAY = 10.0 # Maximum retry delay in seconds
# Rate limiting configuration
_REQUEST_DELAY = 0.5 # Minimum delay before each request (seconds)
_MAX_CONCURRENT_REQUESTS = 3 # Maximum concurrent requests
# Global semaphore for rate limiting
_request_semaphore = asyncio.Semaphore(_MAX_CONCURRENT_REQUESTS)
async def edge_tts(
@@ -28,7 +37,7 @@ async def edge_tts(
pitch: str = "+0Hz",
output_path: str = None,
retry_count: int = _RETRY_COUNT,
retry_delay: float = _RETRY_DELAY,
retry_base_delay: float = _RETRY_BASE_DELAY,
) -> bytes:
"""
Convert text to speech using Microsoft Edge TTS
@@ -38,8 +47,9 @@ async def edge_tts(
Returns audio data as bytes (MP3 format).
Includes automatic retry mechanism to handle 401 authentication errors
and temporary network issues (default: 3 retries with 2s delay).
Includes automatic retry mechanism with exponential backoff and jitter
to handle 401 authentication errors and temporary network issues.
Also includes concurrent request limiting and rate limiting.
Args:
text: Text to convert to speech
@@ -48,8 +58,8 @@ async def edge_tts(
volume: Speech volume (e.g., +0%, +50%, -20%)
pitch: Speech pitch (e.g., +0Hz, +10Hz, -5Hz)
output_path: Optional output file path to save audio
retry_count: Number of retries on failure (default: 3)
retry_delay: Delay between retries in seconds (default: 2.0)
retry_count: Number of retries on failure (default: 5)
retry_base_delay: Base delay for exponential backoff (default: 1.0s)
Returns:
Audio data as bytes (MP3 format)
@@ -74,14 +84,27 @@ async def edge_tts(
"""
logger.debug(f"Calling Edge TTS with voice: {voice}, rate: {rate}, retry_count: {retry_count}")
last_error = None
# Retry loop
for attempt in range(retry_count + 1): # +1 because first attempt is not a retry
try:
if attempt > 0:
logger.info(f"🔄 Retrying Edge TTS (attempt {attempt + 1}/{retry_count + 1}) after {retry_delay}s delay...")
await asyncio.sleep(retry_delay)
# Use semaphore to limit concurrent requests
async with _request_semaphore:
# Add a small random delay before each request to avoid rate limiting
pre_delay = _REQUEST_DELAY + random.uniform(0, 0.3)
logger.debug(f"Waiting {pre_delay:.2f}s before request (rate limiting)")
await asyncio.sleep(pre_delay)
last_error = None
# Retry loop
for attempt in range(retry_count + 1): # +1 because first attempt is not a retry
try:
if attempt > 0:
# Exponential backoff with jitter
# delay = base * (2 ^ attempt) + random jitter
exponential_delay = retry_base_delay * (2 ** (attempt - 1))
jitter = random.uniform(0, retry_base_delay)
retry_delay = min(exponential_delay + jitter, _MAX_RETRY_DELAY)
logger.info(f"🔄 Retrying Edge TTS (attempt {attempt + 1}/{retry_count + 1}) after {retry_delay:.2f}s delay...")
await asyncio.sleep(retry_delay)
# Monkey patch ssl.create_default_context if SSL verification is disabled
if not _SSL_VERIFY_ENABLED:
@@ -134,28 +157,36 @@ async def edge_tts(
if not _SSL_VERIFY_ENABLED:
ssl.create_default_context = original_create_default_context
except (WSServerHandshakeError, ClientResponseError) as e:
# Network/authentication errors - retry
last_error = e
error_code = getattr(e, 'status', 'unknown')
logger.warning(f"⚠️ Edge TTS error (attempt {attempt + 1}/{retry_count + 1}): {error_code} - {e}")
except (WSServerHandshakeError, ClientResponseError) as e:
# Network/authentication errors - retry
last_error = e
error_code = getattr(e, 'status', 'unknown')
error_msg = str(e)
# Log more detailed information for 401 errors
if error_code == 401 or '401' in error_msg:
logger.warning(f"⚠️ Edge TTS 401 Authentication Error (attempt {attempt + 1}/{retry_count + 1})")
logger.debug(f"Error details: {error_msg}")
logger.debug(f"This is usually caused by rate limiting. Will retry with exponential backoff...")
else:
logger.warning(f"⚠️ Edge TTS error (attempt {attempt + 1}/{retry_count + 1}): {error_code} - {e}")
if attempt >= retry_count:
# Last attempt failed
logger.error(f"❌ All {retry_count + 1} attempts failed. Last error: {error_code}")
raise
# Otherwise, continue to next retry
if attempt >= retry_count:
# Last attempt failed
logger.error(f"❌ All {retry_count + 1} attempts failed. Giving up.")
except Exception as e:
# Other errors - don't retry, raise immediately
logger.error(f"Edge TTS error (non-retryable): {type(e).__name__} - {e}")
raise
# Otherwise, continue to next retry
except Exception as e:
# Other errors - don't retry, raise immediately
logger.error(f"Edge TTS error (non-retryable): {e}")
raise
# Should not reach here, but just in case
if last_error:
raise last_error
else:
raise RuntimeError("Edge TTS failed without error (unexpected)")
# Should not reach here, but just in case
if last_error:
raise last_error
else:
raise RuntimeError("Edge TTS failed without error (unexpected)")
def get_audio_duration(audio_path: str) -> float:
@@ -184,20 +215,20 @@ def get_audio_duration(audio_path: str) -> float:
return max(1.0, estimated_duration) # At least 1 second
async def list_voices(locale: str = None, retry_count: int = _RETRY_COUNT, retry_delay: float = _RETRY_DELAY) -> list[str]:
async def list_voices(locale: str = None, retry_count: int = _RETRY_COUNT, retry_base_delay: float = _RETRY_BASE_DELAY) -> list[str]:
"""
List all available voices for Edge TTS
Returns a list of voice IDs (ShortName).
Optionally filter by locale.
Includes automatic retry mechanism to handle network errors
(default: 3 retries with 2s delay).
Includes automatic retry mechanism with exponential backoff and jitter
to handle network errors and rate limiting.
Args:
locale: Filter by locale (e.g., zh-CN, en-US, ja-JP)
retry_count: Number of retries on failure (default: 3)
retry_delay: Delay between retries in seconds (default: 2.0)
retry_count: Number of retries on failure (default: 5)
retry_base_delay: Base delay for exponential backoff (default: 1.0s)
Returns:
List of voice IDs
@@ -213,14 +244,26 @@ async def list_voices(locale: str = None, retry_count: int = _RETRY_COUNT, retry
"""
logger.debug(f"Fetching Edge TTS voices, locale filter: {locale}, retry_count: {retry_count}")
last_error = None
# Retry loop
for attempt in range(retry_count + 1):
try:
if attempt > 0:
logger.info(f"🔄 Retrying list voices (attempt {attempt + 1}/{retry_count + 1}) after {retry_delay}s delay...")
await asyncio.sleep(retry_delay)
# Use semaphore to limit concurrent requests
async with _request_semaphore:
# Add a small random delay before each request to avoid rate limiting
pre_delay = _REQUEST_DELAY + random.uniform(0, 0.3)
logger.debug(f"Waiting {pre_delay:.2f}s before request (rate limiting)")
await asyncio.sleep(pre_delay)
last_error = None
# Retry loop
for attempt in range(retry_count + 1):
try:
if attempt > 0:
# Exponential backoff with jitter
exponential_delay = retry_base_delay * (2 ** (attempt - 1))
jitter = random.uniform(0, retry_base_delay)
retry_delay = min(exponential_delay + jitter, _MAX_RETRY_DELAY)
logger.info(f"🔄 Retrying list voices (attempt {attempt + 1}/{retry_count + 1}) after {retry_delay:.2f}s delay...")
await asyncio.sleep(retry_delay)
# Monkey patch SSL if verification is disabled
if not _SSL_VERIFY_ENABLED:
@@ -258,24 +301,32 @@ async def list_voices(locale: str = None, retry_count: int = _RETRY_COUNT, retry
if not _SSL_VERIFY_ENABLED:
ssl.create_default_context = original_create_default_context
except (WSServerHandshakeError, ClientResponseError) as e:
# Network/authentication errors - retry
last_error = e
error_code = getattr(e, 'status', 'unknown')
logger.warning(f"⚠️ List voices error (attempt {attempt + 1}/{retry_count + 1}): {error_code} - {e}")
except (WSServerHandshakeError, ClientResponseError) as e:
# Network/authentication errors - retry
last_error = e
error_code = getattr(e, 'status', 'unknown')
error_msg = str(e)
# Log more detailed information for 401 errors
if error_code == 401 or '401' in error_msg:
logger.warning(f"⚠️ Edge TTS 401 Authentication Error (list_voices attempt {attempt + 1}/{retry_count + 1})")
logger.debug(f"Error details: {error_msg}")
logger.debug(f"This is usually caused by rate limiting. Will retry with exponential backoff...")
else:
logger.warning(f"⚠️ List voices error (attempt {attempt + 1}/{retry_count + 1}): {error_code} - {e}")
if attempt >= retry_count:
logger.error(f"❌ All {retry_count + 1} attempts failed. Last error: {error_code}")
raise
if attempt >= retry_count:
logger.error(f"❌ All {retry_count + 1} attempts failed. Giving up.")
except Exception as e:
# Other errors - don't retry, raise immediately
logger.error(f"List voices error (non-retryable): {type(e).__name__} - {e}")
raise
except Exception as e:
# Other errors - don't retry, raise immediately
logger.error(f"List voices error (non-retryable): {e}")
raise
# Should not reach here, but just in case
if last_error:
raise last_error
else:
raise RuntimeError("List voices failed without error (unexpected)")
# Should not reach here, but just in case
if last_error:
raise last_error
else:
raise RuntimeError("List voices failed without error (unexpected)")