平台介绍Platform Overview

将文本转换为自然语音。支持多种音色和语言，可调节语速、音调等参数，适用于智能客服、有声读物、语音导航等场景。Convert text to natural speech. Supports multiple voices and languages with adjustable speed and pitch, ideal for virtual assistants, audiobooks, and voice navigation.

支持流式分段情感描述和非流式整段情感描述。流式场景持续发送 PCM 并按有效语音段返回结果；非流式场景上传完整 WAV 并一次性返回最终情感标签或自由文本描述。Supports both streaming segmented emotion captioning and non-streaming whole-file emotion captioning. Streaming sessions send PCM continuously and receive segment results; non-streaming requests upload one WAV file and receive one final label or caption.

高并发情感识别服务，返回 Top-K 情绪标签与置信度，支持队列限流和微批处理。High-concurrency speech emotion recognition service returning Top-K labels and scores with queue-based backpressure and micro-batching.

面向普通中英实时转写，公网入口为 wss://amphion.top/asr/v1/stream，支持用户级热词池和会话临时热词，不包含增强型清洗或翻译后处理。Standard Chinese-English realtime transcription at wss://amphion.top/asr/v1/stream, with user hotword pools and per-session hotwords.

面向术语敏感型实时转写场景，在流式 ASR 上增加行业热词增强、轻量文本清洗、文本情感增强和翻译模式，支持语种指定、内置行业词包、自定义热词和清洗强度控制。Built for terminology-sensitive live transcription, this streaming ASR endpoint adds industry hotword support, lightweight text cleanup, optional text emotion, and translation mode, with language hints, built-in term packs, custom hotwords, and cleanup-level controls.

面向录音文件、会议回放和课堂音频，上传整段音频后执行离线识别，并可结合行业热词、文本清洗或翻译模式返回最终结果。For recordings, meeting replays, and classroom audio. Upload a full file for offline ASR with optional industry hotwords, text cleanup, or translation mode.

语音到语音的流式同声传译服务。客户端通过 WebSocket 持续发送 PCM 音频，服务端按稳定片段执行识别、翻译与语音合成，持续返回原文、译文和翻译语音。Streaming speech-to-speech interpretation over WebSocket. Clients send PCM audio continuously; the server performs ASR, translation, and TTS on stable segments, then streams back source text, translated text, and translated audio.

实时语音转文字翻译服务。客户端发送 PCM 音频流，服务端同步返回识别原文和翻译译文（纯文本），支持自动语言检测和多目标语言，端到端延迟通常 < 500ms，适用于实时字幕、跨语言会议、同声传译辅助等场景。Real-time voice-to-text translation. Stream PCM audio and receive both recognized text and translated text with end-to-end latency typically under 500ms. Supports auto language detection and multiple targets, ideal for live subtitles, multilingual meetings, and interpretation assistance.