What the data for this category looks like
Speech annotation and transcription data sources include pending audio files, annotation platform backend data, custom term lists, and manual spot-check samples. Data is updated either daily or per annotation batch. Associated daily documents include transcription specification documents, audio delivery lists, quality inspection reports, and speaker annotation forms. Data fields include audio duration (in seconds), speaker ID, transcribed text (in characters), confidence scores, and there are no unified fixed statistical percentage requirements.
What constraints these characteristics impose on the "HTTP Interfaces and External Systems" link
Diverse audio file formats require interfaces to support ingestion and processing of multiple mainstream audio formats. Batch update requirements require interfaces to support batch calls to avoid overloading single requests. The need to associate multiple document fields requires structured data returned by the interface to match the field requirements of the annotation workflow, reducing subsequent data cleaning work. Time-consuming steps such as noise segment judgment and speaker differentiation require the interface to set a sufficiently long timeout period to avoid mid-process interruptions. Transcription errors can lead to training deviations or delivery delays, requiring the interface to have retry and exception verification mechanisms to reduce failure risk.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Authentication method | Bearer <annotation platform API key> | Complies with the authentication specifications of the annotation platform's HTTP interface, prevents unauthorized requests from being blocked |
| Number of audio files per single request | First 5 entries | Matches the single-processing limit of the annotation platform's batch interface, prevents request timeouts |
| Timeout period | 600 seconds | Covers the processing duration of single long audio transcription and noise segment judgment, adapts to business processing time |
| Number of retries | 3 times | Addresses network fluctuations or temporary annotation platform failures, reduces transcription failure rate |
| Return field configuration | Set responseData to ["transcript_text", "speaker_labels", "confidence_scores"] | Matches the field requirements of quality inspection reports and speaker annotation forms, reduces subsequent data organization work |
| Variable configuration | Set variables to {"audio_format": ["wav", "mp3"], "max_audio_size": "1000 MB"} | Supports mainstream audio formats supported by the annotation platform, matches the platform's file size limits |
This page's configuration instructions apply to FastGPT v4.17.0 and later versions.
Three common mistakes
- Phenomenon: Using curl to call the interface returns a 401 Unauthorized error. The Docker-deployed test tool verifies normally but the interface call fails. Cause: The
Authorizationrequest header is not configured correctly, or the used API key has not been granted speech annotation and transcription permissions. - Phenomenon: When calling the interface in batches, some requests return no results after sequential execution. Cause: The number of audio files in a single request exceeds the concurrency limit of the annotation platform's interface, and requests are not split into batches.
- Phenomenon: The returned results after calling the interface lack the speaker label field. Cause: The
variablesparameter is not set in the request to specify the fields to return, or the returned content is not filtered viaresponseData.
How to confirm the configuration is complete
- Call the FastGPT test interface, check that the returned HTTP status code is 200 OK, confirm that the authentication configuration takes effect.
- Upload a single test audio file, check that the returned results include transcribed text, speaker labels, and confidence scores, matching the configured return fields.
- Initiate a batch request, check that the number of returned results matches the number of requested audio files, confirm that the batch configuration is correct.
- Simulate a network interruption and then retry the request, check that the interface automatically recovers and returns valid results, confirm that the retry count configuration takes effect.
The values provided are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-10-19.