AI_TRANSCRIBE¶

Syntax¶

AI_TRANSCRIBE( <audio_file> [ , <options> ] [, <return_error_details> ] )

Arguments¶

Required:

audio_file

A FILE type object representing an audio file. Use TO_FILE function to create a reference to your staged file.

Optional:

options

An OBJECT value containing zero or more of the following fields.

• timestamp_granularity: A string specifying the desired timestamp granularity. Possible values are:

  • "word": The file is transcribed as a series of words, each with its own timestamp.

  • "speaker": The file is transcribed as a series of conversational “turns,” each with its own timestamp and speaker label.

  If this field is not specified, the entire file is transcribed as a single segment without timestamps by default.

return_error_details

A BOOLEAN flag that indicates whether to return error details in case of error. When set to TRUE, the function returns an OBJECT that contains the value and the error message, one of which is NULL depending on whether the function succeeded or failed. See Error behavior for details.

Returns¶

An string containing a JSON representation of the transcription result. The JSON object contains the following fields:

• "audio_duration": The total duration of the audio file in seconds.

• "text": The transcription of the complete audio file, provided when the timestamp_granularity field is not specified.

• "segments": An array of segments, provided when the timestamp_granularity field is set to "word" or "speaker". Each segment is a JSON object containing the following fields:

  • "start": The start time of the segment in seconds.

  • "end": The end time of the segment in seconds.

  • "text": The transcription text for the segment.

  • "speaker_label": The label of the speaker for the segment, provided when the timestamp_granularity field is set to speaker. Labels are of the form “SPEAKER_00”, “SPEAKER_01”, etc. and are assigned in the order speakers are detected in the audio file.

Error behavior¶

By default, if AI_TRANSCRIBE can’t process the input, the function returns NULL. If the query processes multiple rows, rows with errors return NULL and don’t prevent the query from completing.

The return value on error depends on the return_error_details argument. The following table shows the return value based on the return_error_details argument:

return_error_details	Return value	Description
FALSE Not passed	NULL
TRUE	OBJECT with value and error fields	value: A VARCHAR value containing the transcription result, or NULL if an error occurred. error: A VARCHAR value that contains the error message if an error occurred, or NULL if the function succeeded.

See moreShow less

Expand

For more information about error handling for AI functions, see Snowflake Cortex AI Function: Multirow error handling improvements.

Access control requirements¶

Users must use a role that has been granted the SNOWFLAKE.CORTEX_USER database role. See Cortex LLM privileges for more information on this role.

Usage notes¶

• For a list of supported languages, see Supported languages

  Supported languages are automatically detected. A file can contain multiple languages, each of which is recognized and transcribed. For accurate language detection, speech must begin within the first five seconds of the file.

• AI_TRANSCRIBE supports the following audio and video file formats:

  Audio	FLAC, MP3, MP4, OGG, WAV, WEBM
  Video	MKV, MP4, OGV, WEBM

  See moreShow less

  Expand

  Video files must contain at least one audio track in FLAC, MP3, OPUS, VORBIS, or WAV format.

  Factors such as sample rate, bit depth, and number of channels do not affect transcription, but they might make the file too large to process if they are too high. Internally, AI_TRANSCRIBE uses monophonic audio at 16 KHz, and resamples input files when they are not already in this format

• The maximum audio file size is 700 MB.

• The maximum audio file duration is 60 minutes when timestamp granularity is set to “word” or “speaker”. If timestamp granularity is not used, the maximum duration is 120 minutes.

Examples¶

For examples, see AI Audio examples.

Troubleshooting¶

If the function fails, it raises an error. Common error messages include:

Regional availability¶

AI_TRANSCRIBE is available in the following regions:

• AWS US West 2 (Oregon)

• AWS US East 1 (N. Virginia)

• AWS EU Central 1 (Frankfurt)

• Azure East US 2 (Virginia)

Legal notices¶

Refer to Snowflake AI and ML.

Limitations¶

Snowflake Cortex functions do not support dynamic tables.