swarmauri_llm_whisper 0.11.0.dev1

Provider-specific Swarmauri import package for Whisper transcription, translation, async, and batch speech-to-text workflows.

Swarmauri Whisper Transcription

swarmauri_llm_whisper provides the provider-specific Swarmauri import package for WhisperLargeModel. Despite the older llm package name, the runtime is a speech-to-text adapter that calls Hugging Face Inference for openai/whisper-large-v3 and supports both transcription and translation workflows.

The adapter targets Hugging Face Inference at https://api-inference.huggingface.co/models/openai/whisper-large-v3, accepts local audio file paths, and returns text output for single-file, async, and batch processing workflows.

Why Use This Package?

• Keep Whisper-specific imports explicit in Swarmauri applications.
• Transcribe or translate audio through a Swarmauri component instead of hand-writing Hugging Face Inference requests.
• Reuse one adapter for synchronous, asynchronous, and batch audio processing.
• Bridge older llms import patterns while newer stt-oriented package paths continue to evolve.

FAQ

What does swarmauri_llm_whisper install?

It installs WhisperLargeModel under swarmauri.llms.

Is this a chat LLM package?

No. The runtime is a speech-to-text adapter for openai/whisper-large-v3 on Hugging Face Inference.

Which tasks are supported?

The adapter supports transcription and translation.

Which model is supported today?

The current runtime allowlist contains openai/whisper-large-v3.

Does it support streaming?

No. stream and astream are explicitly unimplemented for this adapter.

What credentials are required?

You need a Hugging Face access token with permission to use the target inference surface.

Features

• WhisperLargeModel for transcription and translation against Hugging Face Inference.
• Sync and async single-file processing.
• Batch and async batch audio workflows.
• Explicit task selection between transcription and translation.
• Compatibility with Python 3.10, 3.11, 3.12, 3.13, and 3.14.

Installation

uv add swarmauri_llm_whisper

pip install swarmauri_llm_whisper

Usage

Set HF_API_KEY in your environment before creating the model.

Transcription

import os

from swarmauri_llm_whisper import WhisperLargeModel

model = WhisperLargeModel(api_key=os.environ["HF_API_KEY"])
text = model.predict("tests/static/test.mp3", task="transcription")

print(text)

Translation

import os

from swarmauri_llm_whisper import WhisperLargeModel

model = WhisperLargeModel(api_key=os.environ["HF_API_KEY"])
text = model.predict("tests/static/test_fr.mp3", task="translation")

print(text)

Async Batch Processing

import asyncio
import os

from swarmauri_llm_whisper import WhisperLargeModel


async def main() -> None:
    model = WhisperLargeModel(api_key=os.environ["HF_API_KEY"])
    results = await model.abatch(
        {
            "tests/static/test.mp3": "transcription",
            "tests/static/test_fr.mp3": "translation",
        }
    )
    print(results)


# asyncio.run(main())

Examples

• Use transcription when the output should stay in the original spoken language.
• Use translation when the output should be translated into English.
• Use batch methods when one job needs to process multiple audio files together.

Related Packages

• swarmauri_llm_openai
• swarmauri_llm_playht
• swarmauri_llm_groq
• swarmauri_llm_perplexity
• swarmauri_llm_hyperbolic
• swarmauri_llm_llamacpp

Foundational Swarmauri Packages

• swarmauri
• swarmauri_core
• swarmauri_base
• swarmauri_standard

More Documentation

• LLM provider model and pricing links
• openai/whisper-large-v3 model card
• Hugging Face Inference Providers

Best Practices

• Keep HF_API_KEY in environment variables or a secret manager.
• Choose translation only when English output is the intended downstream behavior.
• Use explicit local audio file paths and validate the file exists before dispatching a batch job.
• Prefer newer Swarmauri STT-native imports when available for new projects, since this package preserves an older compatibility surface.

License

Apache-2.0