genblaze-hume

Hume AI Octave TTS provider adapter for genblaze.

Octave is an LLM-based text-to-speech model. This connector exposes it as a synchronous audio provider — generate speech, get a provenance manifest, and persist the result to Backblaze B2 / S3 like any other genblaze step.

Install

pip install genblaze-hume          # or: pip install "genblaze[hume]"
export HUME_API_KEY="..."          # from https://platform.hume.ai/

Usage

from genblaze_core import Modality, Pipeline
from genblaze_hume import HumeTTSProvider

run, manifest = (
    Pipeline("narration")
    .step(
        HumeTTSProvider(output_dir="output/"),
        model="octave-2",
        prompt="Welcome to the future of media provenance.",
        modality=Modality.AUDIO,
        voice_id="Ava Song",          # a Hume Voice Library voice name
        output_format="mp3",          # mp3 | wav | pcm
    )
    .run()
)

print(manifest.canonical_hash)

step.model selects the Octave version: octave-2 → API version="2" (multi-language, requires a voice), octave-1 → version="1".

Pricing

The SDK ships no hardcoded prices. Octave bills per character of input text; register a recipe at runtime — see docs/reference/pricing-recipes.md ("Hume" section):

from genblaze_core.providers import per_input_chars

provider = HumeTTSProvider(api_key="...")
provider.models.register_pricing("octave-2", per_input_chars(RATE, per=1000))

Docs

• Octave TTS overview
• synthesize-json reference
• genblaze new-provider guide