feat: add OpenAI audio transcription proxy node

[collindutter]

Adds an OpenAiAudioTranscription node that transcribes audio to text using OpenAI's GPT-4o transcription and Whisper models via the Griptape Cloud proxy.

The node supports three models: gpt-4o-transcribe, gpt-4o-mini-transcribe, and whisper-1. It accepts audio input in common formats (mp3, mp4, wav, flac, etc.) and outputs the transcribed text. When using verbose_json response format, the node also outputs word-level and segment-level timing data, detected language, and audio duration.

Key parameters include language (ISO-639-1 code to improve accuracy), prompt (context to guide transcription style), response_format (json or verbose_json), and temperature (sampling randomness). Advanced parameters are collapsed by default.

Depends on the proxy client PR: https://github.com/griptape-ai/griptape-cloud/pull/TBD

Sources

• https://platform.openai.com/docs/api-reference/audio/createTranscription
• griptape-ai/griptape-nodes-app#42

Testing with the engine

To test the node end-to-end in the Griptape Nodes UI:

1. Check out both branches:
   • This repo: feat/openai-audio-transcription-proxy-node
   • griptape-cloud: the proxy client branch (see linked PR above)
2. Start the local griptape-cloud: cd griptape-cloud && make up/debug
3. Create DB records for the model config (the migration 0003_add_dummy_default_models handles this automatically)
4. Start the engine with proxy overrides:

   GT_CLOUD_PROXY_BASE_URL=http://localhost:8000 GT_CLOUD_PROXY_API_KEY=local make run/watch

5. In the Nodes UI, add an OpenAI Audio Transcription node and connect it to a workflow
6. Provide an audio file and run the workflow
7. Verify the transcribed text appears in the node's output

Integration test

To run the automated integration test:

GT_CLOUD_PROXY_BASE_URL=http://localhost:8000 GT_CLOUD_PROXY_API_KEY=local \
  uv run python tests/integration/test_openai_audio_transcription.py --language en

Closes griptape-ai/griptape-nodes-app#42

[shhlife]

@collindutter instead of a new node, can we just replace the existing transcribe audio node?