Skip to content

Releases

2026.11.1

Choose a tag to compare

@hwdsl2 hwdsl2 released this 16 Nov 14:19

Adds optional WHISPER_API_KEY_FILE support for loading an API key from a mounted file. Management commands report the key loaded at startup, while existing saved keys are preserved.

Other changes

  • The Colab demo now exposes WHISPER_DIARIZE_THREADS, which inherits WHISPER_THREADS by default. Guidance is clearer on CPU diarization, startup downloads and rerunning the configuration and server cells.

Updating

Existing setups can update normally with their current configuration and volumes. Mounted secret files are optional; no credential migration is required.

If you opt into mounted secret files, use an image with secret-file support and mount a readable credential file. Remove the matching direct variable from the container configuration and any mounted env file, even if its value is empty. Recreate the affected container with its existing volumes, following the linked configuration guide.

With file-based authentication, keys are not copied into persistent key files. Helpers use a private startup snapshot, which needs writable runtime storage accessible to the service user. Back up the external secret source separately. Before disabling file mode, configure another credential source; otherwise an existing volume without a direct or saved key can resume legacy unauthenticated operation.

When rotating a file-sourced credential, make the new contents visible, restart the affected container and refresh consumers that cache the credential. Recreate the container if mounts or environment settings change, or if replacing a bind-mounted file leaves the old contents visible. Before rolling back a file-based setup, ensure the older image can use a supported direct or saved credential source; it will ignore the new file inputs.

Validation

Image builds, authentication tests and Docker Compose secret-mount checks passed for the tested variants. CUDA-image checks ran without a GPU; GPU acceleration was not tested. Podman, rootless operation, Swarm and Kubernetes deployments were not tested. Build and test results.

This is a source and documentation release. Docker images update independently through their existing channels; this source tag does not pin an image or provide a tested image rollback target.

Secret-file setup | Changelog | Changes since 2026.10.3

2026.10.3

Choose a tag to compare

@hwdsl2 hwdsl2 released this 31 Oct 20:58

This release adds configurable speaker embedding models for diarization and improves model downloads, caching and startup.

Improvements

  • Choose from 21 supported models with WHISPER_DIARIZE_EMBEDDING_MODEL. WeSpeaker ResNet34-LM and NeMo TitaNet large are alternatives for English audio. Compare results on your own recordings and retune the clustering threshold after switching.

  • Configure the embedding model and clustering threshold in the Colab notebook.

  • Pre-download another supported embedding without changing the active model:

    docker exec whisper whisper_manage --downloaddiarize wespeaker_en_voxceleb_resnet34_LM.onnx
  • Docker and Colab startup now allow more time for diarization model downloads and retries. The whisper_manage --downloadmodel command now returns a nonzero exit status when a download fails.

Upgrade notes

The default embedding model, cache filenames and API response formats remain unchanged. Diarization still requires WHISPER_DIARIZATION=true, runs on the CPU and applies only to non-streaming requests. Invalid embedding filenames stop startup before model loading, even when diarization is disabled.

  • Cached diarization models are now checked by file size and SHA-256. Valid files, including symlinked models, are reused. Invalid files are preserved and identified in the logs; move the affected cache entries aside and restart with downloads enabled. If you previously swapped weights under another model's filename, restore the official file or rename the replacement to its correct supported filename and select it explicitly. If a diarization model cannot be verified, downloaded or loaded, diarization is disabled while transcription remains available.
  • Offline deployments: WHISPER_LOCAL_ONLY now also blocks automatic diarization downloads. Pre-cache both the segmentation model and the selected embedding model. Explicit whisper_manage --downloaddiarize prefetch remains permitted.
  • Manual script updates: use api_server.py, diarizer.py and the new diarization_models.py from the same revision. Docker images and the notebook include the helper.

See the diarization guide for model selection and recovery details.

Thanks to @surfnterd for configurable embedding model support in #16.

This release identifies a source and documentation snapshot. Docker images continue to update independently through the moving :latest and :cuda tags.

Changelog · Full comparison

2026.10.2

Choose a tag to compare

@hwdsl2 hwdsl2 released this 28 Oct 03:57

This release rejects unsupported turbo translation requests, clarifies model selection, and changes the Colab demo default to multilingual small.

Compatibility

POST /v1/audio/translations now returns HTTP 400 when the loaded model is large-v3-turbo or turbo, for both batch and SSE requests. Turbo models are not trained for speech-to-English translation; previously, these requests were accepted and could produce source-language text or unreliable translation.

For translation, configure server-side WHISPER_MODEL to a multilingual non-turbo model such as base, small, medium, or large-v3. Restart the container for a bind-mounted env file, or recreate it for Docker/Compose environment changes. The request's model field does not switch the loaded model.

Turbo transcription and the Docker default of base are unchanged.

Improvements

  • Default the Colab demo to multilingual small, with clearer model-selection guidance.
  • Correct API error messages, endpoint documentation, localized READMEs, CLI model-list help, and configuration comments.
  • Expand API regression coverage for model restrictions, authentication, batch/SSE behavior, and continued transcription and multilingual translation support.

Docker images continue to update through the existing moving tags; this release identifies the tagged source revision.

Changelog · Full comparison

2026.10.1

Choose a tag to compare

@hwdsl2 hwdsl2 released this 24 Oct 06:21

First tagged source release of ScribeCrate, a transcription and English translation API powered by faster-whisper.

This release includes:

  • OpenAI-compatible transcription and translation, subtitles, word timestamps, and streaming results, with optional speaker diarization for non-streaming responses.
  • CPU and NVIDIA CUDA deployment, persistent model caching, and API-key and model-management commands.

Speaker diarization runs on CPU and inherits WHISPER_THREADS unless WHISPER_DIARIZE_THREADS is set. Set WHISPER_DIARIZE_THREADS=1 to retain the earlier setting of one thread per model. New installations with persistent storage generate an API key by default. Existing installations without a configured key retain their previous behavior.

Documentation | Changelog

This release records a source and documentation snapshot. Docker tags such as latest continue to update independently.