Overview
This guide provides an overview of the schema used for Nuclei Speech Recognition. This schema can be used by integration partners that desire to receive JSON formatted transcripts from Nuclei to enable custom post-processing, including:
- Word-level timestamps
- Word-level accuracy confidence scores
- Speaker segmented transcripts
Example use cases include:
- Select a word from transcript and jump to the correct audio segment for replay.
- Hover over a word from transcript and see the accuracy / confidence score for that word.
- Display speaker segmented transcript summaries.
Schema
Each JSON transcript contains the following additional metadata:
- id
- metadata
- identify_language
- identify_multiple_languages
- language_code
- language_options
- media_format
- settings
- channel_identification
- max_speaker_labels
- show_speaker_labels
- results
- summary
- items
- start_time
- end_time
- alternatives
- confidence
- content
- type
Examples
The following example demonstrates the transcript schema for a basic recording:
{
"id": "c70e570f-c970-49af-a90c-4bfeb76b6f95",
"metadata": {
"identify_language": true,
"identify_multiple_languages": false,
"identified_language_score": 0.9849987030029297,
"language_code": "en-US",
"language_options": ["de-DE", "en-US", "fr-FR"],
"media_format": "mp3",
"settings": {
"channel_identification": false,
"max_speaker_labels": 2,
"show_speaker_labels": false
}
},
"results": {
"summary": "Hello, world!",
"items": [
{
"start_time": "0.64",
"end_time": "1.24",
"alternatives": [
{
"confidence": "1.0",
"content": "Hello"
}
],
"type": "pronunciation"
},
{
"alternatives": [
{
"confidence": "0.0",
"content": ","
}
],
"type": "punctuation"
},
{
"start_time": "1.39",
"end_time": "1.71",
"alternatives": [
{
"confidence": "1.0",
"content": "world"
}
],
"type": "pronunciation"
},
{
"alternatives": [
{
"confidence": "0.0",
"content": "!"
}
],
"type": "punctuation"
}
]
}
}
More Information
For more information, please contact Nuclei's sales team at hello@nuclei.ai
Comments
0 comments
Article is closed for comments.