Fish Speech S2 Pro: Local Installation and Creative Use

Text-to-Speech (TTS) has made giant leaps, but often the best models are locked behind expensive APIs or restrictive licenses. Fish Speech S2 Pro breaks this mold, offering state-of-the-art performance with fully open-source model weights.

In this guide, we will see how to install S2 Pro locally and start experimenting with voice cloning and emotional control.

Why S2 Pro Is Different

S2 Pro doesn't just read text; it supports real-time prosody and emotion control through natural-language tags. Want the model to whisper? Add [whisper]. Want it to laugh? Use [laugh].

Local Installation

Installation requires a GPU with decent VRAM (8GB-12GB+ recommended for optimal performance).

1. Clone the repository:

   git clone https://github.com/fishaudio/fish-speech
   cd fish-speech

2. Install dependencies:

   pip install -r requirements.txt

3. Download model weights: Follow the instructions in the repository to download S2 Pro checkpoints.

Voice Cloning and Expressiveness

The true power of S2 Pro lies in its ability to clone voices with just a few seconds of reference audio and overlay complex emotions.

# Semantic generation example
python tools/tts/generate_semantic.py \
    --text "Welcome to my channel. Today we are testing Fish Speech S2 Pro [laugh]." \
    --ref ref_voice.npy \
    --output code_0.spk

Creative Use

The possibilities are endless: from automated podcasts with realistic cloned voices to virtual assistants with a defined emotional personality. Ethics are paramount here: always use voice cloning only with the explicit consent of the voice owners.

Conclusion

Fish Speech S2 Pro is an incredible tool for anyone who wants to explore the future of open-source voice generation. Installing it locally gives you full control over your audio data without depending on any third-party provider.

Have you tried cloning a voice yet? How did the emotional expression quality turn out?