close

FAQ.

FAQ

What is the maximum duration of audio I can transcribe?

If you are POSTing an audio file to /v1/transcript, there is no limit on the size or duration of audio you can send to the API.

Using /v1/stream you can transcribe up to 15 seconds of speech for an immediate transcript.

What sample rate should the audio be?

If you are POSTing an audio file to /v1/transcript, there is no required format. You can send any sample rate in any format (mp3, wav, flacc, etc) and we'll turn the audio into text.

Using /v1/stream you must send live audio in 8khz sample rate, mono (single channel), with 16bit little-endian formatting.

Can I perform signal processing on the audio before I send it to the API?

You shouldn't perform any signal processing on the audio you send to the API. Doing so will have a negative impact on the accuracy in most cases. While the processed audio may sound cleaner to you, our neural network will be confused by it.

How many phrases or words can I add to a Corpus?

At this time, you can send a maximum of 5000-7000 phrases/words, depending on the length of each phrase/word. If this is too restrictive for you, please send an email to support@assemblyai.com.

If I set the "closed_domain" flag to True, will the API be able to recognize phrases/sentences not in the examples file?

Yes. The closed_domain flag only restricts the vocabulary of the API. For example, if your phrases include: "hello world", and "hi", the API will still be able to recognize "hi world".

Troubleshooting

If something went wrong, the response status code will be in the 400 range and you should have an "error": <message> in your response.

Error messages usually fall into three categories:

What do I do if I get an error?

For any response with "status": "error" please refer to "error": <message>. Here are some common error messages:

Error message Explanation
Corpus creation error, invalid parameter: {attribute} Please double-check the json parameters.
Download error, 404 Client Error: Not Found for url Please double-check that the provided url is valid.
Audio conversion error, could not decode. Please double-check that the url returns a playable audio file. Internally we normalize audio to wav files using ffmpeg.
Server error, developers have been alerted. We will resubmit these requests as soon as possible.

How can I get help or learn more?

Send yourself an invite to join our Slack community.

Or send an email to: