Get Episode Transcripts

Example:

{
  getEpisodeTranscript(uuid:"e03bf3ef-829e-4f47-9f02-29ac6a747b4f"){
    id
    text
    speaker
    startTimecode
    endTimecode
  }
}

{
  getPodcastEpisode(uuid:"e03bf3ef-829e-4f47-9f02-29ac6a747b4f"){
    uuid
    name
    taddyTranscribeStatus
    transcript
    transcriptWithSpeakersAndTimecodes{
      id
      text
      speaker
      startTimecode
      endTimecode
    }
  }
}

curl -H "X-USER-ID: 1" -H "X-API-KEY: xyz..." \                                                                                           
https://ax2.taddy.org/9d874d17-fe25-4cbb-a802-ce65f7c198a1/106d1dfb-50ed-4844-8e04-31960d8767c7/transcript.vtt

How Taddy API generates transcripts

1. Podcast-provided transcripts - Some podcasts provide their own text transcripts (however, less than 1% of podcasts currently do this)

2. Automatic transcription for popular podcasts - We automatically generate transcripts for the most popular 5000 podcasts using an open-source transcription model running on our GPUs

3. On-demand transcription - We can transcribe any episode on-demand (It takes ~10 seconds to transcribe every ~1hr of audio)

Pricing

• 100 episode transcripts/month are included on the Pro Plan

• 2000 episode transcripts/month are included on the Business Plan

• Need additional episode transcripts? Get an additional 2000 episode transcripts/month for $100/month (5c per transcript)

• If an episode provides its own text transcript with timecodes i.e.) if a transcript is provided in WEBVTT (.vtt) or SubRip (.srt) format, we use that transcript and do not generate our own transcript. This transcript is available to all Taddy API users (Free and Paid) and does not count against your episode transcripts/month.

• You can request the same transcript multiple times in the same month and it will only count once. We keep track of all the transcripts that a user has requested for that month and reset it when your billing cycle resets every month.

How to check if a podcast is one of the top 5000 podcasts we are automatically transcribing every episode of:

{
  getPodcastSeries(name:"This American Life"){
    uuid
    name
    taddyTranscribeStatus
  }
}

How to check if a transcript exists for an episode:

{
  getPodcastEpisode(uuid:"e03bf3ef-829e-4f47-9f02-29ac6a747b4f"){
    uuid
    name
    taddyTranscribeStatus
  }
}

Try our transcription feature for Free

{
  getPodcastSeries(uuid:"6bdfd429-f58b-427d-8072-353d478aa15f") {
    uuid
    name
    rssUrl
    episodes(limitPerPage:1){
      uuid
      name
      datePublished
      audioUrl
      duration
      taddyTranscribeStatus
      transcript
    }
  }
}

Searching only for episodes that have a transcript

{
  search(term:"Neil deGrasse Tyson", filterForTypes:PODCASTEPISODE, sortBy:POPULARITY, filterForHasTranscript:true){
    searchId
    podcastEpisodes{
      uuid
      name
      datePublished
      description
      audioUrl
      transcript
    }
  }
}

See how many episode transcripts you have left for the month

{
  getTranscriptCreditsRemaining
}

Useful GraphQL properties

" Status of transcript (complete, processing, not transcribed) "
taddyTranscribeStatus: PodcastEpisodeTranscriptionStatus

" Downloads the transcript, parses it and returns an array of text in paragraphs. "
transcript: [String]

" Download the transcript, parses it and return an array of transcript items (which includes text, speakers and timecodes) "
transcriptWithSpeakersAndTimecodes(
  " (Optional) Style option for transcript. Default is PARAGRAPH"
  style: TranscriptItemStyle
): [TranscriptItem]

" A list of urls where you can download the transcript for this episode "
transcriptUrls: [String]

" A list of urls where you can download the transcript for this episode, including more details "
transcriptUrlsWithDetails: [TranscriptLink]

Technical Details

• The most popular 5000 podcasts were picked based on the Most Popular Podcasts query.

• We do not attempt to identify individual speakers in the transcript or infer who is speaking. If your use case requires this, you can explore speaker diarization (segmenting audio by speaker) and speaker identification (identifying the name of each speaker) and can use our transcript as input.

• For on-demand transcript, which are created as Taddy API users request them, we use a faster open-source model that is within a 1% word error rate of our standard transcription model.

• Some podcasts use dynamic ad insertion. ie) The same audio link will give different listeners different ads depending on location and other factors. This will affect the timestamps provided in transcripts as there may be different ads in the audio file we transcribed.

• We use queues to prioritize which episodes we transcribe. Newly released episodes are high priority and transcribed as soon as possible. We're also working through the entire back catalogue of the most popular 5000 podcasts, which is a lower priority.

• Transcripts automatically generated by Taddy API are Brotli encoded. This only affects you if you're accessing a transcript provided by us directly via its URL. Most libraries will automatically decode Brotli, but if you're using a library that doesn't, be sure to decode it yourself.

Referenced types in this document: