Speaker Diarization

Learn how to detect multiple speakers in an audio

The Speaker Diarization model lets you detect multiple speakers in an audio file and what each speaker said.

If you enable Speaker Diarization, the resulting transcript will return a list of utterances, where each utterance corresponds to an uninterrupted segment of speech from a single speaker.

Speaker Diarization and multichannel

Speaker Diarization doesn’t support multichannel transcription. Enabling both Speaker Diarization and multichannel will result in an error.

Quickstart

To enable Speaker Diarization, set speaker_labels to true in the transcription config.

Python

import assemblyai as aai

aai.settings.api_key = "YOUR_API_KEY"

# You can use a local filepath:
# audio_file = "./example.mp3"

# Or use a publicly-accessible URL:
audio_file = (
"https://assembly.ai/wildfires.mp3"
)

config = aai.TranscriptionConfig(
speaker_labels=True,
)

transcript = aai.Transcriber().transcribe(audio_file, config)

for utterance in transcript.utterances:
print(f"Speaker {utterance.speaker}: {utterance.text}")

Example output

Speaker A: Smoke from hundreds of wildfires in Canada is triggering air quality alerts throughout the US. Skylines from Maine to Maryland to Minnesota are gray and smoggy. And in some places, the air quality warnings include the warning to stay inside. We wanted to better understand what's happening here and why, so we called Peter DiCarlo, an associate professor in the Department of Environmental Health and Engineering at Johns Hopkins University. Good morning, professor.
Speaker B: Good morning.
Speaker A: So what is it about the conditions right now that have caused this round of wildfires to affect so many people so far away?
Speaker B: Well, there's a couple of things. The season has been pretty dry already, and then the fact that we're getting hit in the US. Is because there's a couple of weather systems that are essentially channeling the smoke from those Canadian wildfires through Pennsylvania into the Mid Atlantic and the Northeast and kind of just dropping the smoke there.
Speaker A: So what is it in this haze that makes it harmful? And I'm assuming it is.
...

Set number of speakers

If you know the number of speakers in advance, you can improve the diarization performance by setting the speakers_expected parameter.

Python

config = aai.TranscriptionConfig(
speaker_labels=True,
speakers_expected=3
)

The speakers_expected parameter is ignored for audio files with a duration less than 2 minutes.

API reference

Request

curl https://api.assemblyai.com/v2/transcript \
--header "Authorization: YOUR_API_KEY" \
--header "Content-Type: application/json" \
--data '{
  "audio_url": "YOUR_AUDIO_URL",
  "speaker_labels": true,
  "speakers_expected": 3
}'

Key	Type	Description
speaker_labels	boolean	Enable Speaker Diarization.
speaker_expected	number	Set number of speakers.

Response

{utterances:[...]}

Key	Type	Description
utterances	array	A turn-by-turn temporal sequence of the transcript, where the i-th element is an object containing information about the i-th utterance in the audio file.
utterances[i].confidence	number	The confidence score for the transcript of this utterance.
utterances[i].end	number	The ending time, in milliseconds, of the utterance in the audio file.
utterances[i].speaker	string	The speaker of this utterance, where each speaker is assigned a sequential capital letter. For example, “A” for Speaker A, “B” for Speaker B, and so on.
utterances[i].start	number	The starting time, in milliseconds, of the utterance in the audio file.
utterances[i].text	string	The transcript for this utterance.
utterances[i].words	array	A sequential array for the words in the transcript, where the j-th element is an object containing information about the j-th word in the utterance.
utterances[i].words[j].text	string	The text of the j-th word in the i-th utterance.
utterances[i].words[j].start	number	The starting time for when the j-th word is spoken in the i-th utterance, in milliseconds.
utterances[i].words[j].end	number	The ending time for when the j-th word is spoken in the i-th utterance, in milliseconds.
utterances[i].words[j].confidence	number	The confidence score for the transcript of the j-th word in the i-th utterance.
utterances[i].words[j].speaker	string	The speaker who uttered the j-th word in the i-th utterance.

The response also includes the request parameters used to generate the transcript.

Frequently asked questions

How can I improve the performance of the Speaker Diarization model?

To improve the performance of the Speaker Diarization model, it’s recommended to ensure that each speaker speaks for at least 30 seconds uninterrupted. Avoiding scenarios where a person only speaks a few short phrases like “Yeah”, “Right”, or “Sounds good” can also help. If possible, avoiding cross-talking can also improve performance.

How many speakers can the model handle?

The upper limit on the number of speakers for Speaker Diarization is 10.

How accurate is the Speaker Diarization model?

The accuracy of the Speaker Diarization model depends on several factors, including the quality of the audio, the number of speakers, and the length of the audio file. Ensuring that each speaker speaks for at least 30 seconds uninterrupted and avoiding scenarios where a person only speaks a few short phrases can improve accuracy. However, it’s important to note that the model isn’t perfect and may make mistakes, especially in more challenging scenarios.

Troubleshooting

Why is the speaker diarization not performing as expected?

The speaker diarization may be performing poorly if a speaker only speaks once or infrequently throughout the audio file. Additionally, if the speaker speaks in short or single-word utterances, the model may struggle to create separate clusters for each speaker. Lastly, if the speakers sound similar, there may be difficulties in accurately identifying and separating them. Background noise, cross-talk, or an echo may also cause issues.