Make whisper an optional extra with faster-whisper by default

[Dreamsorcerer]

Problem

Whisper requires downloading a 150MB model and depends on torch (with GBs of CUDA downloads).

Solution

Provide faster-whisper by default (2MB) and use as a fallback when whisper is not available.
This avoids the 150MB download, and means we are one step closer to not depending on torch for a base install.

Breaking Changes

Users need to request dimos[whisper] now for full whisper feature.

Test

python -c "
from dimos.stream.audio.pipelines import stt
node = stt()
node.emit_text().subscribe(on_next=lambda t: print(f"USER: {t}"))
from dimos.stream.audio.utils import keepalive
keepalive()
"

[Dreamsorcerer]

TTS seems to work pretty well with faster-whisper anyway.

[greptile-apps]

Greptile Summary

This PR makes openai-whisper optional by introducing faster-whisper as the default backend (bundled in the agents extra), while moving openai-whisper to the dev extra under a new dimos[whisper] install path.

• node_whisper.py now has a two-stage import try/except at module level: prefers openai-whisper if present, falls back to faster-whisper, and raises a clear ImportError if neither is found.
• pyproject.toml replaces openai-whisper with faster-whisper>=1.0.0 in the agents extra and adds openai-whisper to dev; faster_whisper is also added to the mypy ignore list.

Confidence Score: 4/5

Safe to merge after fixing the misleading install hint in the ImportError message.

The ImportError raised when neither backend is found tells users to run pip install dimos[whisper], but that extra does not exist in pyproject.toml — openai-whisper lives only in dev. Any user who hits this error and follows the hint will get a pip failure instead of a working install. The rest of the change — backend selection logic, compute_type mapping, segment joining, and dependency pinning — looks correct.

dimos/stream/audio/stt/node_whisper.py — specifically the ImportError message at the bottom of the fallback try/except block.

Important Files Changed

Filename	Overview
dimos/stream/audio/stt/node_whisper.py	Added module-level try/except for backend selection; transcription path split per backend. Contains a misleading install hint in the ImportError message and a deprecated logger.warn call.
pyproject.toml	Swaps openai-whisper for faster-whisper>=1.0.0 in the agents extra and adds openai-whisper to dev; no whisper extra is defined for end users who want the full backend.
uv.lock	Lock file updated to include faster-whisper 1.2.1 and its ctranslate2 dependency; openai-whisper moved to dev group. Generated file, no logic to review.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Module import] --> B{import whisper}
    B -- success --> C[_USE_FASTER_WHISPER = False]
    B -- ImportError --> D{import faster_whisper.WhisperModel}
    D -- success --> E[_USE_FASTER_WHISPER = True\nlog warning]
    D -- ImportError --> F[raise ImportError\n'No whisper backend found']

    C --> G[WhisperNode.__init__]
    E --> G

    G --> H{_USE_FASTER_WHISPER?}
    H -- True --> I[WhisperModel\ndevice=auto\ncompute_type from fp16]
    H -- False --> J[whisper.load_model]

    I --> K[emit_text → transcribe\njoin segments]
    J --> L[emit_text → transcribe\nresult text strip]

Loading

Comments Outside Diff (1)

1. dimos/stream/audio/stt/node_whisper.py, line 43-47 (link)

   The error message tells users to run pip install dimos[whisper], but no whisper extra exists in pyproject.toml — openai-whisper was moved to dev only. Running that command will fail with "invalid extra". The hint should direct users to install openai-whisper directly or via dimos[dev].

_{Reviews (2): Last reviewed commit: "Merge branch 'dev' into sam/move-whisper" | Re-trigger Greptile}