Speech API process overview

Upload audio, video, or transcripts for processing in Medallia Speech.

When using the Speech API, you can upload raw audio or video files to be transcribed by Speech, or you can upload complete transcripts that were created by your own Automatic Speech Recognition (ASR) engine. Connectors do not support ingesting complete transcripts at this time.

Process raw audio files

When Medallia Speech ingests audio via the Speech API, we create a transcript in JSON format containing utterances and attributes that are used by Text Analytics and Reporting. All metrics and insights are calculated from that transcript rather than directly from the audio.

To process audio or video files via the Speech API to create a feedback record in Experience Cloud, follow the steps below:

  1. Upload an audio or video file to your Medallia Media File Transfer location, and note the full path to the file.
  2. Send a POST request to the /v0/bulk-ingest Speech API endpoint (see the Supported file types section for details).

    This endpoint is used to provide context for voice signals by sending metadata associated with the signals (audio files) in the body of the request.

    The request must include the full path to your file, including the file name.

When the request is received, Medallia Speech will process the audio, calculate and map metrics to Experience Cloud fields, make the transcript available for Text Analytics, and populate reports.

Upload complete transcripts

If you wish to provide your own complete JSON transcript rather than raw audio or video, you must be careful to format the transcript exactly as the Speech pipeline would have formatted it, including all necessary attributes, so that services further along the pipeline can work with the data.

To upload an existing transcript from your own ASR engine via the Speech API, follow the steps below:

  1. Create a JSON transcript in the required format.
    For detailed information about the required format, see ASR transcript schema.
  2. Upload the transcript to your Medallia Media File Transfer location, and note the full path to the file.
  3. Send a POST request to the /v1/bulk-ingest-transcription Speech API endpoint (see the Supported file types section for details).

    The request must include the full path to your transcript, including the file name.

When the request is received, Medallia Speech will process the transcript, calculate and map metrics to Experience Cloud fields, make the transcript available for Text Analytics, and populate reports.

Supported file types

Format audio files to be processed by Medallia Speech.

To ensure that audio files contain usable data that can be processed by Medallia Speech to provide meaningful insights, requirements are in place for filenames, audio formats, and audio duration.

Audio formats and filenames

Filenames must adhere to the following guidelines:

  • Alphanumeric characters only

  • Maximum of 512 characters in length for the path and filename combined

    For example, /folder/subfolder/file.mp3

  • No special characters/symbols

Speech supports the following audio formats for transcription:

  • FLAC
  • AAC
  • Opus
  • MP3
  • WAV
  • AVI
  • OGG_OPUS
  • M4A
  • MP4
  • PGP/PGP

Lossless compression formats (such as FLAC) are best, because they optimize for network bandwidth when transferring files to Medallia Experience Cloud. If you use lossy compression formats (such as AAC and MP3), ensure the compression is tuned to not disrupt audio recordings more than 2%. Lossless formats are preferred.

Tip:

While Experience Cloud supports all formats listed above, we recommend using the following formats (in order of preference):

  • FLAC
  • AAC
  • Opus
  • MP3
  • WAV

Files cannot be zipped. Files can be encrypted with a dedicated Speech PGP key. For more information, see the following guides:

Note: If you do not see your preferred file type listed in this guide, contact a Medallia expert to discuss alternatives or the possibility of supporting a new file type.

Audio duration

To ensure that audio is neither too short to contain a meaningful interaction or too long to process efficiently, Speech enforces the following minimum and maximum duration:

EngineMinimum durationMaximum duration
Engine 1 (Medallia's ASR)2 seconds6 hours (21,600 seconds)
Engine 2 (Amazon Transcribe)2 seconds4 hours (14,400 seconds)

An audio file that is shorter or longer than the minimum/maximum duration will be considered invalid and will not be processed.