Chinese to English Voice Translator — Mandarin & Cantonese Audio to Text

Upload Chinese audio (MP3, WAV, M4A) or video (MP4, MOV). Auto-detect handles Mandarin (Whisper Tier 1, ~8% WER on FLEURS-ZH). CANTONESE — Whisper trained primarily on Mandarin; Cantonese output treated as first-draft. Set target to English. Export as TXT, DOCX, SRT, VTT. Output defaults to English text; character output for reverse direction (English → Chinese) defaults to Simplified. Free 30 min on signup.

With honest caveats. Whisper Large-v3 was trained primarily on Mandarin (Standard Chinese) — Cantonese (广东话) is a separate spoken variety with distinct phonology and tonal system. Output is usable for content understanding and first-draft translation, but expect noticeably lower accuracy than Mandarin. For broadcast-grade Cantonese work, use a Cantonese-specialized transcription service. Practical recommendation: if your source speaker can switch to Mandarin (which most educated Cantonese speakers can in business/media contexts), record in Mandarin for best quality.

For reverse direction (English → Chinese), output defaults to Simplified Chinese (used in Mainland China, Singapore). Traditional Chinese (used in Taiwan, Hong Kong, Macau) available as post-processing option. For Chinese → English direction, output is English (Simplified/Traditional character choice doesn't apply since output is English text).

Tier 1 on clean Mandarin audio. Whisper Large-v3 on FLEURS-ZH: ~8% WER on clean broadcast Mandarin. Neural translation on Chinese↔English: 20-30 BLEU on standard WMT test sets. Combined pipeline: ~82-88% word-level accuracy on clean Mandarin audio. Drops on: tonal clarity issues (background music, phone call quality, room ambience) — Mandarin's 4-tone system requires clean audio; regional accents (Beijing, Shanghai, Taiwan) with variance; music-heavy audio; multiple simultaneous speakers.

Yes. WeChat is the dominant messaging app in Chinese-speaking world and Chinese diaspora communities. WeChat exports voice notes as .amr or .m4a files. Long-press the voice message → Save (varies by WeChat client version). Upload to VexaScribe — source auto-detects as Chinese, target = English. Common workflow: Chinese diaspora processing voice notes from family in Mainland China / Taiwan / Hong Kong.

Tier 3 (draft-quality only). Shanghainese (上海话), Hokkien / Taiwanese Hokkien (闽南语/台语), Sichuanese (四川话), and other regional Chinese varieties are separate spoken languages from Mandarin — Whisper Large-v3 was not specifically trained on these. Output is usable for rough understanding but treat as draft only. Most speakers of these varieties also speak Mandarin — if possible, record in Mandarin for reliable transcription.

Yes. Reverse direction: upload English audio, source auto-detects as English, target = Chinese (Simplified default; Traditional available). Output uses standard written Chinese. Common workflow: English brand video → Simplified Chinese for Mainland China market; English content → Traditional Chinese for Taiwan / Hong Kong market.

Yes on the free tier — 30 minutes of transcription + translation on signup, no credit card. Translation included on all plans. Premium tier (higher-accuracy internal model) available at 1 credit per 5,000 characters for client-facing content.

MP3, WAV, M4A, OGG, FLAC, AAC, AMR, OPUS all supported. WeChat .amr and .m4a exports work directly. Chinese business audio commonly comes as MP3 or M4A from smartphone recordings; conference calls as WAV or M4A. Max 5 GB per file.

Yes significantly. Mandarin uses 4 tones (plus neutral tone) — tonal distinctions carry lexical meaning. Poor audio quality (background music, phone compression, room echo, distant microphone) obscures tone contours and degrades ASR accuracy. Expect 5-15% WER increase on tonally-unclear audio. Best quality: close-mic studio recording or clean broadcast audio.

Yes. Upload MP4, MOV, WebM, MKV — VexaScribe extracts audio, transcribes Chinese, translates to English. Export SRT for English subtitles on original Chinese video. Common use case: Chinese YouTube / Bilibili / Douyin content localization for English audiences, C-drama fan subtitle work, Chinese business video content research.

Upload the recording (usually MP3 or M4A from phone/Zoom/Teams), source auto-detects as Chinese (Mandarin), target = English. Speaker diarization identifies participants across the call. Export as DOCX with speaker labels for meeting minutes, or SRT if you want timestamped review. Common use case: sourcing/manufacturing/tech calls with Mainland China contacts, cross-border partnership discussions.