Speech to Text - 90+ Languages, Word Timestamps

Speech to text, but more than "audio in, words out"

Speech to text turns the words spoken in an audio or video file into a transcript you can edit, search and publish. This tool adds three things on top of plain dictation:

• 90+ languages, auto-detected — no language picker, mixed-language content handled
• Speaker diarization — up to 32 speakers, every word labeled with who said it
• Word-level timestamps — exact start/end per word, one click to SRT subtitles

All three come back from a single upload. There is no second pass to time the text and no third pass to work out who said what.

A real transcription

We ran the Chinese homepage sample through it. Below is the raw, unedited output:

中国語7秒

Numbers and terminology in a lecture recording are the most direct test of transcription accuracy:

ボイス: 知性御姐中国語16秒

Detected: zho (confidence 98.3%)
Text: 三步做出爆款口播，念帖文案，选择音色，一键到处带字幕的成品。
Word timestamps (first 6):
三 [0.20s→0.44s]  步 [0.44s→0.68s]  做 [0.68s→0.88s]
出 [0.88s→1.06s]  爆 [1.06s→1.28s]  款 [1.28s→1.40s]

Full disclosure: two homophones were misheard in this clip ("粘贴"→"念帖", "导出"→"到处") — homophones remain the frontier of speech recognition, whoever builds the engine. That's exactly why we show raw output: you can edit the text in the studio and the timestamps stay intact.

So the realistic expectation is this: most words are right on the first pass, while homophones, names and jargon need one read-through from you.

How to convert speech to text in three steps

1. Open Voice Studio → Transcribe and upload audio or video — up to 200MB per file.
2. Skip the language picker. The language is detected automatically, and diarization is on by default, so there is nothing to configure.
3. Read the result once, fix any wrong words, then copy the text or download the SRT from the history panel.

Step 3 has one rule worth remembering: correct first, export second. Editing a word never moves its timestamps, so the transcript and the subtitles only need proofreading once.

Typical uses

• Subtitling: transcribe video → download SRT → drop into your editor. The full subtitle workflow is on the subtitle generator page.
• Interviews & meetings: diarization outputs "who said what" directly. Multi-person recordings are covered in meeting transcription.
• Podcast show notes: one episode, one click, one SEO-friendly transcript.
• Video into articles: search engines cannot hear speech, so a transcript is the first draft of a written piece.
• Voiceover QA: transcribe generated audio back to verify script fidelity.

Looking for something more specific? Audio files: MP3 to text. Video files: MP4 to text. YouTube videos: YouTube transcript generator.

Speech to text vs forced alignment vs typing it yourself

Typing a transcript by hand is the slowest option of all: an hour of audio means endless pausing and rewinding. The efficient split is to let the machine draft and keep the human for proofreading.

Habits that improve accuracy

• Record close to the microphone. Switch off constant noise such as fans or air conditioning — the cleaner the audio, the fewer wrong words.
• Give each speaker their own mic in multi-person recordings. Diarization gets steadier, and cross-talk is easier to separate.
• Proofread proper nouns first. Names, brands and technical terms are what homophones replace most often.
• Lock a video edit before transcribing. Re-cut the video afterwards and the SRT no longer matches the new picture.

FAQ

Do I need to pick the language before converting speech to text? No. The language is detected automatically across 90+ languages, and mixed-language content is handled without a language picker.

Can it tell who is speaking? Yes. Speaker diarization is on by default for up to 32 speakers, and every word is labeled with who said it, so interviews and meetings come back as a dialogue transcript.

What if the transcript has mistakes? Homophones are a hard limit for every speech recognition engine — the real sample on this page has two. Edit the text in the studio; the word-level timestamps stay intact, so you can export the corrected SRT straight away.

How much does transcription cost? 20 credits per file, up to 200MB per upload, refunded automatically if transcription fails. The $3.90 trial pack has 150 credits — enough for 7 files — and they never expire.

Start here

Cost: 20 credits per file (up to 200MB per upload), refunded automatically if it fails. The $3.90 trial pack has 150 credits, enough for 7 files, and they never expire.

Upload the recording you most need written up and get back an editable speech to text transcript with word-level timestamps.

Transcribe my first file →