Removed hardcoded API limit of 3000 and made it configurable via env plus docker compose files#52
Open
hastla007 wants to merge 79 commits into
Open
Removed hardcoded API limit of 3000 and made it configurable via env plus docker compose files#52hastla007 wants to merge 79 commits into
hastla007 wants to merge 79 commits into
Conversation
Add minimum character requirement for Long Text API
Add minimum length requirement for Long Text async API
…-minimum-length Make long text minimum length configurable
…-minimum-length-scvjtj Fix long text minimum length validation
…-minimum-length-amdgll Ensure long text limits respect runtime configuration
Reduce minimum length for Long Text async API from 1000 to 100 characters.
…lity-presets-32lmip Add configurable quality presets to long text TTS
…ehandler-3y7c8x Align pause defaults with docker configuration
This commit fixes 5 critical bugs found during comprehensive code review: 1. **long_text.py line 156**: Remove incorrect await on non-async get_progress() - get_progress() is a synchronous function, calling it with await caused TypeError - Fixed by removing await keyword 2. **long_text.py line 462**: Remove incorrect await on non-async cancel_job() - cancel_job() is a synchronous function, calling it with await caused TypeError - Fixed by removing await keyword 3. **long_text.py line 468**: Remove incorrect await on non-async delete_job() - delete_job() is a synchronous function, calling it with await caused TypeError - Fixed by removing await keyword 4. **speech.py lines 1060-1124**: Fix temp file cleanup race condition in SSE streaming - Previously, temp file was being deleted by outer finally block before SSE generator finished - This caused file not found errors during streaming audio generation - Fixed by moving cleanup logic: SSE path uses generator's finally, non-SSE uses outer try-finally 5. **speech.py lines 1048, 1249**: Replace bare except with except OSError - Bare except catches all exceptions including SystemExit and KeyboardInterrupt - Fixed to catch only OSError for file cleanup operations - Improves code quality and prevents masking critical exceptions All bugs were found through: - Static code analysis searching for await on non-async functions - Manual code review of resource management patterns - Analysis of exception handling practices Testing: Code changes verified through static analysis. Runtime testing requires full TTS model installation which is not included in this commit.
…ehandler-yrroj6 Standardize pause defaults across docker compose variants
…U4NC42yuz5Ju6 Fix critical API bugs in long_text and speech endpoints
Fixed 4 critical bugs in app/api/endpoints/speech.py: 1. Lambda closure bugs (3 instances): - Lines 595-611: generate_speech_streaming() - Lambda captured loop variables by reference - Lines 806-822: generate_speech_sse() - Lambda captured loop variables by reference - Replaced lambdas with proper function factories to capture variables by value - Impact: Prevents incorrect TTS parameters being used during concurrent processing 2. Uninitialized variable bugs (3 instances): - Lines 198, 513, 720: initial_memory variable only initialized when ENABLE_MEMORY_MONITORING=true - Added explicit initialization to None before conditional assignment - Changed locals() check to None check (line 456) - Impact: Prevents NameError when memory monitoring is disabled All fixes verified with syntax checking.
…SuysQgFPp2RpCS832hTc Fix critical bugs in speech endpoints
This commit fixes 5 critical bugs across the codebase: Bug #1 (app/main.py): HTTPException detail format inconsistency - Fixed exception handler to ensure consistent error response format - Now handles both dict and string detail values properly - Wraps string details in proper {"error": {...}} structure Bug #2 (app/main.py): Model initialization task cancellation error - Fixed variable scope issue with model_init_task - Stored task in app.state for access during shutdown - Prevents NameError when attempting to cancel task Bug #3 (app/core/text_processing.py): Empty list validation - Added validation to prevent concatenation of empty audio chunks list - Raises ValueError with clear error message - Prevents IndexError when accessing audio_chunks[0] Bug #4 (app/core/audio_processing.py): Audio normalization division by zero - Fixed potential division by zero in _normalize_audio_levels - Now filters segments with valid dBFS before calculating average - Returns original segments unchanged if no valid dBFS values exist Bug #5 (app/api/endpoints/speech.py): Variable scope issue in cleanup - Initialized final_audio_cpu variable at function start - Prevents NameError in finally block cleanup code - Ensures proper cleanup regardless of execution path All fixes have been tested and verified with unit tests.
…s-01YDhomHJWPFXitrarqL36NV Fix critical bugs in audio processing and error handling
- Add /v1/languages alias for /languages endpoint - Add /v1/ping alias for /ping endpoint - Ensures consistent API access through both primary and /v1 prefixed paths - Improves OpenAI API compatibility
…LDzQVUpnCuqG8 Fix missing endpoint aliases for /languages and /ping
The PERIOD_PAUSE_MS environment variable was missing from .env.example but was present in .env.example.docker and referenced in app/config.py and app/core/pause_handler.py. This configuration controls the pause duration (in milliseconds) after periods in text-to-speech generation. Changes: - Added PERIOD_PAUSE_MS=500 to .env.example pause handling section - Ensures consistency between example configuration files - Aligns with existing implementation in Config class and PauseHandler This fixes a configuration inconsistency that could cause confusion for users setting up the application locally using .env.example.
…oQhXSdk8rsFYAiPohKq Fix missing PERIOD_PAUSE_MS configuration in .env.example
Implements comprehensive multi-model architecture allowing users to select
between different Chatterbox TTS model versions via API parameter.
Key Features:
- Support for 4 model variants: chatterbox-v1, chatterbox-v2,
chatterbox-multilingual-v1, chatterbox-multilingual-v2
- Runtime model selection via 'model' parameter in API requests
- Lazy loading of models to optimize memory usage
- OpenAI-compatible model names (tts-1, tts-1-hd) map to default model
- Model registry system for managing multiple loaded models
Changes:
- requirements.txt: Updated to use official chatterbox-tts==0.1.4 package
- app/core/tts_model.py: Complete rewrite to support multi-model registry
- Added load_model() for loading specific model versions
- Added get_or_load_model() for lazy loading
- Added get_available_models() to list all available models
- Updated all getters to support model version parameter
- app/models/requests.py: Added 'model' field to TTSRequest with validation
- app/api/endpoints/speech.py: Updated all TTS endpoints to support model parameter
- Updated generate_speech_internal() to accept model_version
- Updated generate_speech_streaming() to accept model_version
- Updated all endpoint calls to pass model parameter
- app/api/endpoints/models.py: Updated /models endpoint to list all available models
- .env.example: Added DEFAULT_MODEL_VERSION configuration option
API Usage Examples:
# Use V2 multilingual model
POST /v1/audio/speech
{"input": "Hello world", "model": "chatterbox-multilingual-v2"}
# Use V1 model
POST /v1/audio/speech
{"input": "Hello world", "model": "chatterbox-v1"}
# OpenAI compatibility (uses default)
POST /v1/audio/speech
{"input": "Hello world", "model": "tts-1"}
Backward Compatibility:
- If no model parameter is provided, uses default configured in .env
- Default is chatterbox-multilingual-v2 for best results
- Existing API calls continue to work without changes
- Added Model Selection section with comparison table - Included usage examples for all model variants - Documented configuration options - Explained lazy loading behavior - Added OpenAI compatibility notes
…7RSDZsQiuAaCx1rGoQ1s1i Claude/add chatterbox v2 tts 017 rsd zs qiu aa cx1r go q1s1i
Implements comprehensive support for language-specific models from HuggingFace,
enabling automatic downloading and loading of specialized models for 12+ languages.
Features:
- Auto-download from HuggingFace repositories on first use
- Support for .pt and .safetensors model formats
- 12 language-specific models (EN, DE, FR, IT, RU, JA, KO, NO, HY, KA)
- Multiple variants for German (default, havok2, SebastianBodza)
- New API endpoints: /languages, /languages/{code}/models, /language-models
- Lazy loading and caching for efficient memory usage
- Seamless integration with existing model selection system
New modules:
- app/core/language_models.py: Language model configuration and registry
- app/core/model_downloader.py: HuggingFace model download and loading
- app/api/endpoints/language_models.py: Language model API endpoints
- docs/LANGUAGE_MODELS.md: Comprehensive documentation
Updated:
- app/core/tts_model.py: Extended to support language-specific models
- app/api/router.py: Added language models endpoints
- requirements.txt: Added huggingface-hub and safetensors dependencies
- .env.example: Added language models documentation
- README.md: Added language models section with usage examples
Available language models:
- English (ResembleAI/chatterbox)
- German (stlohrey/chatterbox_de, niobures/Chatterbox-TTS)
- French (Thomcles/ChatterBox-fr)
- Italian, Russian, Japanese, Korean, Norwegian, Armenian, Georgian (niobures/Chatterbox-TTS)
Usage:
POST /v1/audio/speech with model="chatterbox-de" for German TTS
GET /languages to list all supported languages
GET /language-models to see all available models
Implements a comprehensive multi-engine TTS architecture allowing users to choose from multiple TTS models: Chatterbox (default), IndexTTS-2, and Higgs Audio V2. New Features: - Multi-engine abstraction layer with BaseTTSEngine interface - IndexTTS-2 support with emotion control and zero-shot voice cloning - Higgs Audio V2 support with multi-speaker and long-form generation - Lazy loading: models download automatically on first use - Per-request model selection via 'model' parameter - Backward compatible with existing Chatterbox models Architecture Changes: - Created app/core/tts_engines/ package with engine implementations: - base.py: BaseTTSEngine abstract interface - chatterbox.py: ChatterboxEngine wrapper - indextts.py: IndexTTSEngine with auto-download - higgs_audio.py: HiggsAudioEngine integration - Refactored app/core/tts_model.py to use engine registry - Extended model registry to support 6 model variants Configuration: - Added DEFAULT_TTS_ENGINE environment variable (chatterbox/indextts/higgs) - Added huggingface-hub dependency for model downloads - Updated .env.example and .env.example.docker with new options Documentation: - Created comprehensive docs/TTS_MODELS.md guide - Includes installation, usage, comparison, and troubleshooting Available Models: 1. chatterbox-v1/v2 (English-only, 1GB, 4-8GB VRAM) 2. chatterbox-multilingual-v1/v2 (23 languages, 1GB, 4-8GB VRAM) - default 3. indextts-2 (emotion control, 1-2GB, 8GB+ VRAM) 4. higgs-audio-v2 (multi-speaker, 3-4GB, 24GB+ VRAM) All models maintain OpenAI API compatibility and support voice cloning.
…-01UNvt1dFVwSAk3GiZLMZhXS Add language-specific ChatterBox TTS models with auto-download support
…D1w622f4p26rwQE Add IndexTTS-2 and Higgs Audio V2 TTS engines with multi-model support
Implements integration for VibeVoice, an expressive long-form conversational speech synthesis system supporting multi-speaker podcasts and dialogues. Features: - Two model variants: VibeVoice-1.5B and VibeVoice-7B - Lazy loading pattern - models load only when first requested - Multi-speaker support (up to 4 speakers) - Long-form generation (up to 90 minutes for 1.5B, 45 minutes for 7B) - Multilingual support (12+ languages) - Zero-shot voice cloning - Context-aware synthesis with LLM-powered understanding Changes: - Created app/core/tts_engines/vibevoice.py: VibeVoiceEngine implementation - Updated app/core/tts_engines/__init__.py: Export VibeVoiceEngine - Updated app/core/tts_model.py: Register vibevoice-1.5b and vibevoice-7b models - Updated requirements.txt: Add VibeVoice installation instructions - Added VIBEVOICE_INTEGRATION.md: Comprehensive integration documentation - Added tests/test_vibevoice_integration.py: Integration test suite - Added verify_vibevoice_integration.py: Verification script Model Details: - vibevoice-1.5b: 64K context, ~90 min generation, ~3GB, 8GB+ VRAM - vibevoice-7b: 32K context, ~45 min generation, ~14GB, 16GB+ VRAM API Usage: - Model IDs: "vibevoice-1.5b" and "vibevoice-7b" - Compatible with all existing API endpoints - Supports OpenAI-compatible /v1/audio/speech endpoint - Auto-downloads models from HuggingFace on first use All verification checks passed. Integration is complete and ready for use.
…zGT8q2wpBeZ7xUu3QTpp Add VibeVoice TTS engine with lazy loading support
Fixed 4 critical bugs that would prevent the API from working: 1. Fixed missing return statement in get_model_info() - Function was building info dict but never returning it - Would cause API endpoints to receive None instead of model metadata 2. Fixed incorrect function call in SSE streaming - Changed get_model() to get_or_load_model() with proper error handling - Added model_version parameter support for OpenAI compatibility 3. Fixed missing model_version parameter in generate_speech_sse() - Added model_version to function signature - Updated all callers to pass the parameter - Prevents NameError when function tries to use the variable 4. Fixed missing model parameter in stream_text_to_speech_with_upload() - Added model parameter to function signature - Allows model selection in streaming upload endpoint - Prevents NameError when calling generate_speech_streaming() All bugs validated with syntax checking and code inspection. See BUGFIX_REPORT.md for detailed analysis. Models affected: VibeVoice, IndexTTS-2, Higgs Audio V2, and all Chatterbox variants
…k1yn9LhpPForXhEE Fix critical bugs in new TTS model integrations
This comprehensive bug fix addresses resource leaks, type inconsistencies, validation gaps, and async/await issues across the codebase. ## Bugs Fixed (13 total) ### Critical Severity (3) - Fix resource leaks in IndexTTS and Higgs Audio temp file cleanup - Fix missing await on async queue operation in long_text_jobs.py - Fix file I/O operations without error handling in voice_library.py ### High Severity (6) - Fix metadata corruption risk with atomic file writes - Fix type mismatch for estimated_completion (float -> datetime) - Fix division by zero risk in text_processing.py - Fix missing file existence check in voice hashing - Fix race condition in voice file cleanup - Fix inconsistent type annotation for duration_seconds ### Medium Severity (4) - Fix missing content-type validation in download endpoint - Fix missing parameter validation in VibeVoice engine - Fix missing minimum chunk size validation (now >= 100 chars) - Improve error messages and logging throughout ## Files Modified Core TTS Engines: - app/core/tts_engines/indextts.py (resource leak fix) - app/core/tts_engines/higgs_audio.py (resource leak fix) - app/core/tts_engines/vibevoice.py (parameter validation) Core Services: - app/core/long_text_jobs.py (async/await fix) - app/core/voice_library.py (file I/O, metadata, cleanup fixes) - app/core/status.py (type mismatch and annotation fixes) - app/core/text_processing.py (division by zero fix) API Endpoints: - app/api/endpoints/long_text.py (content-type validation) Configuration: - app/config.py (chunk size validation) Tests: - tests/test_bugfixes_2025_11_16.py (comprehensive tests for all fixes) Documentation: - BUGFIX_SUMMARY_2025-11-16.md (detailed bug report) - CODEBASE_ANALYSIS.md (complete codebase analysis) - API_ENDPOINTS_QUICK_REFERENCE.md (endpoint documentation) - COMPREHENSIVE_TEST_STRATEGY.md (testing guidelines) - ANALYSIS_SUMMARY.txt (summary of findings) ## Impact - Eliminated resource leaks that could cause disk exhaustion - Prevented metadata corruption through atomic file operations - Fixed race conditions in async operations - Added fail-fast validation for configuration errors - Improved type consistency across codebase - Enhanced error messages for debugging - Maintained full backwards compatibility ## Testing - All Python files validated for syntax errors - Comprehensive test suite added for bug fixes - Integration tests for voice library workflow - Type consistency tests - Validation tests for all engines Fixes #[issue number if applicable]
…1AgYbZjWKQnWSdnRx6NysAU Fix 13 critical bugs and improve test coverage
…odules This commit addresses 8 bugs found through comprehensive code analysis: CRITICAL BUGS FIXED (2): - Add missing logging imports to long_text.py and voice_library.py * Prevents NameError crashes when logger is used at runtime * Files: app/api/endpoints/long_text.py, app/core/voice_library.py HIGH SEVERITY BUGS FIXED (5): - Fix race condition on global REQUEST_COUNTER in speech.py * Added threading.Lock to protect concurrent access * Prevents request count corruption in multi-threaded environments * File: app/api/endpoints/speech.py - Fix memory leak in streaming audio generation * GPU tensors now properly freed before CPU conversion * Applied fix to both regular streaming and SSE streaming functions * File: app/api/endpoints/speech.py - Fix type mismatch in date comparison for job filtering * Added None checks before datetime comparisons * Prevents TypeError when timestamps are missing * File: app/core/long_text_jobs.py - Fix shared state modification without lock in job manager * Added threading.Lock to protect active_jobs dictionary access * Prevents race conditions in pause_job and cancel_job methods * File: app/core/long_text_jobs.py MEDIUM SEVERITY BUGS FIXED (2): - Fix potential division by zero in job details calculation * Made length check more explicit for clarity * File: app/api/endpoints/long_text.py - Fix potential integer overflow in WAV header creation * Added bounds checking to prevent chunk_size overflow * File: app/api/endpoints/speech.py Impact: These fixes improve API stability, prevent crashes, eliminate memory leaks, and resolve concurrency issues that could cause data corruption. Testing: All fixes preserve existing functionality while addressing edge cases and error conditions. Manual testing recommended for streaming endpoints.
…8yUU2QKjT2o2v Fix 8 critical and high-severity bugs across API endpoints and core m…
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Changed the hardcoded limit of long text API, setup new variable LONG_TEXT_MIN_LENGTH=100 in env, Dockerfile and docker compose files. So for automations (like n8n) you can have also now an async API if have less than 3000 characters. I tested it with the limit (now 100) and it works like it should be.
Also changed the Docker commands in Readme to give Docker Compose a name (tts-api) as Docker Desktop would name it just like the folder "Docker".