Help

FAQ: Speech to Text

The questions readers ask about transcribing audio and cleaning recordings in the portal, from accepted files and limits to cost, results and export.

Which audio files can I upload?

MP3, WAV, FLAC, OGG, M4A, AAC or WEBM, by file extension, up to 1 hour long and 150 MB per file; the page sends one file per upload. The badges under the upload area name four of the seven, and all seven are accepted. See Speech to Text formats.

Can I record from my microphone instead?

Yes. Press Record audio, allow the browser to use the microphone, and press Stop & Use to send the take through the same checks as an upload; Cancel discards it and nothing is charged. A take stops on its own at 1 hour and is named recording- followed by a number, with the extension your browser records in. See Transcribe a file.

Why is my one-hour WAV refused when a one-hour MP3 is accepted?

The 150 MB limit is checked on the file as you chose it, whatever its length. A one-hour MP3 is about 57 MB; a one-hour stereo 44.1 kHz WAV is far larger and is refused before upload. Export the recording as MP3, or as 16 kHz mono, before you upload it. See Speech to Text formats.

Why does the upload sit at 0% for a while?

An MP3, M4A, AAC or WEBM file is prepared in the browser before it is sent: decoded, mixed to one channel, resampled to 16 kHz and written as WAV. Uploading your audio file stays at 0% while that runs, and a long file can hold there for a while; WAV, FLAC and OGG files are sent as they are. See Speech to Text formats.

Which language is the transcript in?

Pick the language under the file: Auto, Arabic or English. Auto lets the service decide. That is Arabic today, which also handles mixed Arabic/English and clean English speech. Pick Arabic for Arabic and mixed speech. Pick English only for speech that is all English. Arabic sent as English comes back garbled. See Transcribe a file.

What does Turning your audio into text mean, and how long does it take?

The file has been received; once it passes the checks that run after upload, the credits are taken and the transcript is produced. The hint Usually takes about 20 seconds fits a short clip and a long file takes longer; a transcription still running 10 minutes after the file was accepted is marked failed within the next 5 minutes, with its charge refunded, so split a long recording before you press Try again. See Immediate and queued transcription.

Can I close the dialog or leave the page while a transcription runs?

Closing the dialog changes nothing: during the upload it reopens with the next progress update, and afterwards the page keeps waiting and shows the transcript when it arrives. Leaving or reloading the page is different: the transcription finishes and its credits stay charged, but the transcript goes to Assets and the page starts empty on your next visit. See Immediate and queued transcription.

Is a short clip transcribed immediately?

Quickly, but not in one step: every transcription from the page is accepted first, and the transcript arrives when it is ready. The one-call form exists over the API only, for files of up to 10 minutes. See Immediate and queued transcription and, for code, Synchronous and asynchronous calls.

Does Speaker 2 mean a second person was detected?

Yes. With Identify speakers on, which is the default, the voices in one recording are told apart and numbered in the order they first speak: Speaker 1 is whoever speaks first, Speaker 2 the next new voice. Each row of the transcript shows its speaker and the time it starts at, and a line above the text shows each speaker's talk time. The numbers count voices, not names; a speaker the transcript could not place has no label on that row. Transcripts made with the switch off, and older ones, stay one block with a single time, 0:00. See Read a transcript.

How do I save or edit the transcript?

Open the tab you want on the transcript card, Text, Subtitles (SRT) or JSON, and press Download to save it as TXT, SRT or JSON, or Copy to copy what the tab shows. The file is named after your file with its extension replaced, so interview.mp3 gives interview.txt, interview.srt or interview.json. SRT subtitles start each cue with its speaker, such as [Speaker 1], in the language of the page. The text on the page cannot be edited; select and copy a passage, or make your corrections in the downloaded file. See Read a transcript.

Why is the audio I download a WAV when I uploaded an M4A?

The page transcribes the converted 16 kHz mono WAV copy and keeps that copy, so the player, the download icon under Uploaded audios and Assets all carry it. Keep your original if you need it in its own format. See Speech to Text formats.

How much does a transcription cost?

A transcription costs 100 credits per minute of audio, charged per second with a minimum of 1 second, so 15 minutes cost 1,500 credits. The charge is taken when the file passes the checks after upload and refunded in full if the transcription fails, including when it times out, and a file refused before upload is never charged. See Speech to Text limits and credits.

Where do I see what a transcription cost?

The page shows no per-file price. Watch the credit counter in the sidebar, shown as used out of total with the label Used, before and after, or work it out from the length at 100 credits per minute; Credits remaining in the user menu shows the same balance. See Speech to Text limits and credits.

Why did my transcription fail?

Every failure appears in the dialog headed There is a problem with the uploaded file. Please try again. with the reason under it: the file is over 150 MB or longer than 1 hour, its extension is not one of the seven, the balance is short (Insufficient subscription balance), transcription was unavailable, the audio could not be decoded, or no transcript arrived in time (Job timed out — AI server did not respond). A failure after the file was accepted refunds the charge. Each reason and what to do is on Transcribe a file.

Where do my transcripts go when I leave the page?

To the Assets page, opened from the page header or the sidebar, where every transcript made from the portal in the workspace is listed with the audio that was transcribed; Download there saves the text as transcript- followed by the transcript's identifier and .txt. The Speech to Text page itself starts empty on your next visit. See Assets and history.

What does Sound Cleaning do, and when should I use it?

Sound Cleaning keeps the voice in a recording and drops the background, then lets you switch between Original and Cleaned and download the cleaned file, which is always WAV. Use it before you transcribe a recording, before you hand it to an editor, or before you publish audio that was captured in a poor room. See Clean a recording.

What does Sound Cleaning accept, and what does it cost?

MP3, WAV, FLAC or OGG/Opus, up to 1 hour and 150 MB per file. M4A, AAC and WEBM are accepted by Speech to Text and not here, so export them as MP3 or WAV first. A cleanup costs 1,000 credits per minute of audio, charged per second, taken when the file is accepted and refunded if the run stops. See Sound Cleaning formats, limits and credits.

Where are my cleaned recordings?

Under Recent cleanups on the Sound Cleaning page, which lists the newest 8 cleanups made from the portal in the workspace, with Play and Download on each. Cleanups are not listed on the Assets page. From your own code, POST /v1/sound-cleaning cleans a file in one call (up to 10 minutes and 150 MB) and GET /v1/sound-cleaning/jobs lists the API cleanups. See Clean a recording.