在线文字自由发言
将语音转换为 AI 授权转录文本 。 上传音频文件、 麦克风记录或粘贴 URL. 100+ 语言、 10+ 模式、 98+ 准确性 。
1. 上传语音录音
上传音频或视频文件，粘贴URL，或通过麦克风录音。
2. AI将语音转为文字
从10多种AI模型中选择。包含说话人检测和语言自动识别。
3. 导出您的转录文本
支持6种格式下载。分享带音频播放的转录链接。
语音转文字模型
选择适合您需求的AI模型——或让我们为您选择最佳模型。
支持100多种语言的语音转文字
准备好将语音转为文字了吗？
免费开始 →常见问题
Speech to text (also called speech recognition or ASR) converts spoken audio into written words automatically. STT.ai runs your recording through an AI model that listens to the audio and outputs editable text with timestamps and speaker labels — no typing required.
An acoustic model maps the sound waveform to phonemes, then a language model assembles those into the most likely words and punctuation. STT.ai does this on GPU with models like Whisper Large V3 and NVIDIA Canary, so a one-hour recording is usually done in 2-3 minutes.
Yes. Every visitor gets 600 free minutes per month with no signup required for your first file. Paid plans start at $5/month and add longer files, private transcripts, and priority processing.
On clean speech our best models reach 95-97% accuracy (a 3-5% Word Error Rate on benchmarks). Accuracy drops with background noise, heavy accents, crosstalk, or low-bitrate audio — using a decent microphone and a quiet room makes the biggest difference.
Yes. Speak into your microphone and STT.ai streams the transcript live via the live-transcription tool. You can also upload a finished recording for batch transcription if you don't need it word-by-word as you talk.
STT.ai recognizes 100+ languages and auto-detects the spoken language for most audio. You can also set the language manually for a small accuracy lift, and mixed-language recordings are handled by switching mid-clip.
Yes. Speaker diarization labels each voice (Speaker 1, Speaker 2, ...) and you can rename them in the editor. This works across every supported model and language.
STT.ai accepts 20+ formats including MP3, WAV, M4A, FLAC, OGG, MP4, MKV, MOV, WebM, and AVI. Output to TXT, SRT, VTT, DOCX, JSON, or PDF.
Speech to text transcribes WHAT was said into words; voice recognition (speaker identification) determines WHO said it. STT.ai does both — transcription plus speaker diarization — but the terms describe different tasks.
Yes. Audio is processed and deleted by default. Pro plans add client-side encryption so transcripts are unreadable without your key, even to STT.ai, and your data is never used for model training without explicit opt-in.
Yes. STT.ai has a REST API with Python and Node.js SDKs plus an MCP server for Claude and Cursor. The free API tier includes 100 minutes/month, with per-second billing beyond that.
Yes. Every transcript opens in a built-in editor where you can fix misheard words, rename speakers, adjust timestamps, and add notes. Edits persist across every export format.