Install via ClawdBot CLI:
clawdbot install openlark/fish-speechGrade Fair — based on market validation, documentation quality, package completeness, maintenance status, and authenticity signals.
Calls external URL not in known-safe list
https://huggingface.co/fishaudio/s2-proUses known external API (expected, informational)
arxiv.orgAudited Jun 4, 2026 · audit v1.0
Generated Oct 7, 2026
Publishers and creators can generate studio-quality narration in 80+ languages at 44.1 kHz, using emotion tags like [whisper], [excited], and [pause] to direct performance. Voice cloning from a 10-30 second reference keeps a consistent narrator voice across entire series or seasons.
With SGLang-Omni delivering ~100ms TTFA and RTF 0.195, developers can power low-latency conversational IVR, helpdesk bots, and accessibility readers. Streaming PCM output over the OpenAI-compatible endpoint plugs directly into existing telephony or WebRTC pipelines.
Studios can generate distinct character voices using multi-speaker tags (<|speaker:0|>, <|speaker:1|>) and free-form emotion tags for dialogue scenes. Batch synthesis endpoint accelerates producing thousands of lines for NPCs, localization dubs, and animated shorts.
Training teams can synthesize course narration across many languages from one script, then apply voice cloning to maintain a recognizable instructor persona. Caching batch outputs offline lets organizations deploy consistent audio lessons in low-connectivity regions.
Developers can build screen readers and communication aids that convert text to natural, emotionally expressive speech. Cloned personalized voices offer familiarity for users with speech impairments, while adjustable speed and multi-format output (wav/mp3/pcm) support diverse playback devices.
Teams deploy fishaudio/s2-pro on A800/H200 GPUs using vLLM-Omni or SGLang-Omni and expose an OpenAI-compatible /v1/audio/speech endpoint. They monetize by metering characters, minutes of audio generated, or GPU compute hours through their own billing layer.
A cloud vendor offers a turnkey UI and API for voice cloning, batch synthesis, and emotion-tag editing, handling model hosting, autoscaling, and Docker orchestration for customers. Value comes from convenience, uptime SLAs, and compliance rather than raw model access.
An agency records clean reference audio, fine-tunes with LoRA on Slow AR, and delivers branded voice catalogs plus produced audiobooks, ads, or dubbing. Clients pay for the finished creative output while the agency retains the underlying voice pipeline.
💬 Integration Tip
Deploy with vLLM-Omni for the easiest OpenAI-compatible setup, but use SGLang-Omni when you need streaming latency under 100ms; always supply a clean 10-30 second ref_audio with accurate ref_text to avoid mechanical-sounding clones.
Scored Jun 14, 2026
Transcribe audio files with speaker diarization (who speaks when). Supports 100+ languages, automatic language detection, and timestamps. Use for meetings, interviews, podcasts, or voice messages. Requires AssemblyAI API key.
使用 poocr 库识别发票并导出 Excel。当用户需要识别增值税发票、批量处理发票文件或提取发票信息到 Excel 时调用此技能。
Use when the user has audio or video and wants a timestamped transcript (SRT) in the source language. Routes by source language — Chinese defaults to Volcano...
Create, manage, and deploy ElevenLabs conversational AI agents. Use when the user wants to work with voice agents, list their agents, create new ones, or manage agent configurations.
Voice note transcription and archival for OpenClaw agents. Powered by Deepgram Nova-3. Transcribes audio messages, saves both audio files and text transcript...
Text-to-Speech (TTS) and Speech-to-Text (ASR) using coze-coding-dev-sdk. Returns results directly to stdout.