# fish-speech | Fish Audio | Knowledge Base | Instagit

SOTA Open Source TTS

GitHub Stars: 26.1k

Repository: https://github.com/fishaudio/fish-speech

---

## Articles

### [How to Benchmark TTS Quality and Performance Metrics in Fish-Speech](/fishaudio/fish-speech/how-benchmark-tts-quality-performance-metrics)

Benchmark TTS quality and performance with Fish-Speech evaluation scripts. Measure WER, MOS, RTF, and throughput using Seed-TTS and EmergentTTS-Eval for accurate results.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Implement Custom Text Preprocessing Pipelines in Fish-Speech](/fishaudio/fish-speech/how-implement-custom-text-preprocessing-pipelines)

Learn how to implement custom text preprocessing pipelines in Fish-Speech. Easily modify existing functions or create your own to enhance your audio processing.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Handle Out-of-Memory Errors During Training in Fish-Speech](/fishaudio/fish-speech/how-handle-out-of-memory-errors-during-training)

Learn to handle out-of-memory errors in Fish-Speech training. Use mixed-precision, adjust batch size and max length, and enable gradient checkpointing to optimize memory usage.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Set Up Docker Deployment for Production with Fish Speech](/fishaudio/fish-speech/how-set-up-docker-deployment-production)

Learn how to set up Docker deployment for production with Fish Speech. Deploy production-ready images for web interfaces and REST APIs with CUDA or CPU support and health checks.

- Tags: deployment-guide
- Published: 2026-03-12

### [How to Implement Real-Time Streaming TTS with Fish-Speech](/fishaudio/fish-speech/how-implement-real-time-streaming-tts)

Implement real-time streaming TTS effortlessly with Fish-Speech. Get synthesized WAV audio chunk-by-chunk via our production-ready HTTP API, eliminating wait times for instant audio.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Handle Multilingual TTS Without Phoneme Preprocessing in Fish-Speech](/fishaudio/fish-speech/how-handle-multilingual-tts-without-phoneme-preprocessing)

Discover how Fish-Speech achieves multilingual TTS directly from Unicode text, bypassing phoneme preprocessing for seamless speech synthesis in numerous languages. Explore the unified approach.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Configure the Hydra Training Configuration in Fish-Speech](/fishaudio/fish-speech/how-configure-hydra-training-configuration)

Learn to configure the Hydra training configuration in Fish-Speech using modular YAML files. Easily override parameters via command line or custom experiments without touching Python code.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Debug Common VQ Encoding Errors in Fish-Speech](/fishaudio/fish-speech/how-debug-common-vq-encoding-errors)

Encountering VQ encoding errors in Fish-Speech? Learn how to debug them by checking model instances, audio shapes, sample rates, and tensor integrity for smoother audio processing.

- Tags: how-to-guide
- Published: 2026-03-12

### [How the TextToSemantic Lightning Module Handles Training in Fish Speech](/fishaudio/fish-speech/how-text-to-semantic-lightning-module-handle-training)

Learn how the TextToSemantic Lightning module trains models. Discover its approach to optimizer configuration, loss computation, and LoRA checkpointing for efficient speech model development.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Implement Voice Cloning with Reference Audio in Fish-Speech](/fishaudio/fish-speech/how-implement-voice-cloning-reference-audio)

Implement voice cloning with reference audio using Fish-Speech. Learn how Fish-Speech encodes audio into tokens and injects them into a TTS model for realistic voice generation.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Configure the Fish-Speech API Server with Custom Parameters](/fishaudio/fish-speech/how-configure-api-server-custom-parameters)

Learn to configure the Fish-Speech API server with custom parameters like listen, device, and API key. Easily adjust your API setup for optimal performance.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Quantize the Fish-Speech Model for Efficient Inference](/fishaudio/fish-speech/how-quantize-model-efficient-inference)

Learn how to quantize the Fish-Speech model using INT8 and INT4 for efficient inference. Reduce model size by up to 8x while preserving generation quality.

- Tags: how-to-guide
- Published: 2026-03-12

### [How GRPO Reinforcement Learning Alignment Works in Fish-Speech](/fishaudio/fish-speech/how-grpo-reinforcement-learning-alignment-works)

Explore how GRPO reinforcement learning alignment works in Fish-Speech. Discover how text-to-semantic transformers are aligned using data-filtering models as reward models.

- Tags: deep-dive
- Published: 2026-03-12

### [How to Use Inline Emotion and Prosody Control Tags in Fish-Speech](/fishaudio/fish-speech/how-use-inline-emotion-prosody-control-tags)

Easily control speaking style in Fish Speech synthesis with inline emotion and prosody tags. Insert simple bracketed commands like [laugh] or [whispers] into your text for natural expression.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Handle Long-Context Multi-Turn TTS Generation in Fish-Speech](/fishaudio/fish-speech/how-handle-long-context-multi-turn-tts-generation)

Learn to handle long-context multi-turn TTS generation with Fish-Speech. Our stateful pipeline processes dialogue batches, maintaining context across turns for seamless audio.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Merge LoRA Weights into the Base Fish-Speech Model](/fishaudio/fish-speech/how-merge-lora-weights-base-model)

Learn to merge LoRA weights into the Fish-Speech base model by unioning state dictionaries and triggering weight folding. Create clean inference checkpoints efficiently.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Optimize Inference Latency with SGLang Streaming in Fish-Speech](/fishaudio/fish-speech/how-optimize-inference-latency-sglang-streaming)

Optimize inference latency with SGLang streaming in Fish-Speech. Achieve 100ms time-to-first-audio and RTF below 0.2 on GPUs. Boost your speech AI performance today.

- Tags: how-to-guide
- Published: 2026-03-12

### [How VQManager Handles Reference Audio Encoding in Fish Speech](/fishaudio/fish-speech/how-vqmanager-handle-reference-audio-encoding)

Discover how VQManager encodes reference audio into discrete prompt tokens via dynamic sample rate detection, torchaudio loading, tensor shaping, and DAC VQ-GAN encoding. Learn the four-stage process.

- Tags: internals
- Published: 2026-03-12

### [How to Implement Multi-Speaker Voice Generation with Speaker Tokens in Fish-Speech](/fishaudio/fish-speech/how-implement-multi-speaker-voice-generation-speaker-tokens)

Learn to implement multi-speaker voice generation with speaker tokens in Fish-Speech. Condition acoustic generation on speaker context using inline tokens for seamless voice synthesis.

- Tags: how-to-guide
- Published: 2026-03-12

### [How to Set Up API Authentication with Token-Based Security in Fish Speech](/fishaudio/fish-speech/how-set-up-api-authentication-token-based-security)

Learn how to set up API authentication with token-based security for your Fish Speech API. Secure your data by launching the server with an API key and using the Authorization header for requests.

- Tags: how-to-guide
- Published: 2026-03-12

### [How the RVQ Audio Codec Encodes and Decodes Speech in Fish-Speech](/fishaudio/fish-speech/how-rvq-audio-codec-encode-decode-speech)

Discover how the RVQ audio codec encodes speech by quantizing latents and reconstructs waveforms from indices, offering efficient compression in Fish-Speech.

- Tags: deep-dive
- Published: 2026-03-12

### [How to Configure LoRA Fine-Tuning for the Text2Semantic Model in Fish-Speech](/fishaudio/fish-speech/how-configure-lora-fine-tuning-text2semantic-model)

Configure LoRA fine-tuning for Fish-Speech text2semantic models using Hydra. Easily add low-rank adapters to transformer embeddings and linear projections for parameter-efficient training.

- Tags: how-to-guide
- Published: 2026-03-12

### [How the Dual-Autoregressive Architecture Works in Fish Speech S2: A Deep Dive](/fishaudio/fish-speech/how-dual-autoregressive-architecture-works-fish-speech-s2)

Explore the Dual-Autoregressive architecture in Fish Speech S2. Learn how it separates semantic token prediction from VQ-codebook generation using slow and fast transformers for detailed audio.

- Tags: deep-dive
- Published: 2026-03-12