Parler-TTS Mini v1 (878 M params, Apache-2.0)

Pocket-size text-to-speech that you steer with plain-English prompts.

• Prompt-controlled voices. Describe gender, speed, pitch, background noise or reverb—plus 34 built-in “Jon / Lea / Jenna …” speakers for repeatable tone.

• Spec sheet. 878 M-parameter seq-to-seq model, trained on 45 K hours of audio; spits out 24 kHz WAV straight from the decoder—no extra vocoder.

• Light on hardware. ~1.8 GB VRAM in FP16, <500 MB in 4-bit; real-time even on laptop CPUs. Torch 2 torch.compile, SDPA and batching guides give extra speed-ups.

Why pick it for Norman AI?

Open weights, sub-2 GB footprint and prompt-level style control let us add branded voices, audio previews or edge-device narration without new infra or license headaches.

response = await norman.invoke(
    {
        "model_name": "xtts-v2",
        "inputs": [
            {
                "display_title": "Prompt",
                "data": "A female speaker delivers a slightly expressive and animated speech with a high-pitched voice in a clear audio environment
"
            },
            {
                "display_title": "Prompt",
                "data": "/Users/alice/Desktop/sample_input.aac"
            }
        ]
    }
)