Free Chinese Transcription
Transcribe Chinese audio and video to text with AI. Fast, accurate, and free.
How It Works
- Go to the Free.ai Transcriber
- Upload your Chinese audio or video file
- Our AI automatically detects Chinese and transcribes it
- Download your transcript as text or SRT subtitles
Chinese Transcription Features
- ✓Powered by faster-whisper (MIT licensed)
- ✓Automatic Chinese language detection
- ✓Supports MP3, WAV, MP4, M4A, FLAC, and more
- ✓Timestamps and subtitle export (SRT)
- ✓No file size limits on paid plans
- ✓Private and secure -- files are deleted after processing
Language Details
| Language | Chinese |
| ISO Code | zh |
| AI Model | faster-whisper |
| Price | Free |
More Languages
View All LanguagesFAQ
Whisper large-v3-turbo lands in its top accuracy tier on Chinese — under 7% word error rate on standard benchmarks. In practice that means clean studio audio comes back near-perfect, and conversational audio is usable with minimal cleanup. (Tier A, under 7% word error rate on benchmark sets - we publish honest WER tiers rather than marketing claims.)
Yes - Chinese transcription draws from your daily free token pool first. Audio costs about 50 tokens per minute, so the anonymous daily pool covers a few hours of audio per day. Signed-in accounts get a larger 30,000-token daily pool. Past that, transcription is pay-as-you-go, with token top-ups from $1.
Pass language=zh for Mandarin (the default — simplified or traditional output depending on the source). For Cantonese use language=yue if your audio is Hong Kong / Guangzhou speech; Cantonese transcribed as zh will produce a Mandarin-orthography approximation that loses tones and slang.
MP3, WAV, M4A, FLAC, OGG, OPUS, and WEBM are accepted directly. For video (MP4, MOV, MKV) we extract the audio track server-side before sending it to Whisper - you do not need to convert anything yourself. Same pipeline regardless of source language, including Chinese.
Anonymous uploads cap at roughly 500 MB per file. Signed-in accounts go up to 2 GB. Duration is not a hard limit - long files are chunked automatically (30-second windows with overlap) and stitched back into a single transcript with continuous timestamps. Multi-hour Chinese recordings (podcasts, full lectures, meetings) work fine.
Yes - speaker diarization is on by default for every Chinese transcript. The output is segmented as Speaker 1 / Speaker 2 / Speaker 3 with timestamps, so interviews, panel discussions, and multi-party meetings come back labeled. Diarization runs on a separate model and works the same across all languages we support.
Yes - paste the URL into /transcribe/youtube/ for YouTube or /transcribe/podcast/ for podcast feeds (Apple, Spotify, RSS). We download the audio, run it through Whisper with language=zh, and return the transcript with timestamps and speaker labels. Typical Chinese content: podcasts, lectures, interviews, and long-form YouTube content in Chinese are the most common workloads we see.
Whisper costs about 50 tokens per minute of audio, so a one-hour recording is ~3,000 tokens. Most users never spend anything - the free daily pool of 30,000 tokens covers short clips, voice notes, and one-off podcasts. Beyond it, transcription is pay-as-you-go, with token top-ups from $1.
Yes - both segment-level (every ~10-30 seconds) and word-level timestamps are available. Word-level is the default for VTT/SRT subtitle export so the captions sync line-by-line. On the API set timestamps="word" in the request body. Chinese transcripts are returned in native Han characters (UTF-8) — simplified or traditional depending on the source audio and ISO code.
Yes. POST audio (multipart/form-data, field name "file") to /v1/transcribe/ with language=zh - or omit the language parameter to let Whisper auto-detect. Returns JSON with the transcript, segments, timestamps, and speaker labels. Full reference and SDK snippets at /api/.
Yes - once transcription finishes, click Translate or paste the text into /translate/. Chinese pairs with every other language we support (200+). For meeting minutes pipe the transcript through /summarize/; for dubbing send it to /voice/tts/ to render audio in the target language.
Whisper is trained on 680K hours of noisy real-world audio, so Chinese transcription is robust to background noise, music beds, and phone-quality recordings. Severe clipping or multiple overlapping speakers will still hurt accuracy. If a transcript comes back unusable, email contact@free.ai with the file - we will refund the tokens and look at whether a different engine handles your audio better.