crazygiscool/optimus-piper-voice

Optimus Prime Piper Voice

An unofficial English text-to-speech voice model fine-tuned from Piper's en_US-lessac-medium voice to resemble the Optimus Prime character voice associated with Peter Cullen. It exports as a Piper-compatible ONNX model and uses the VITS architecture.

This project is not affiliated with or endorsed by Peter Cullen, Hasbro, Paramount, or the Transformers franchise rights holders.

Model Details

• Developer: Crazygiscool
• Task: English text-to-speech
• Architecture: Piper VITS
• Base model: speaches-ai/piper-en_US-lessac-medium
• Sample rate: 22,050 Hz
• Phonemizer: eSpeak, en-us
• Artifacts: ONNX model, Piper JSON config, and PyTorch training checkpoint
• License stated for this model: CC BY-NC 4.0

The license above does not grant rights to the Transformers character, any performer's voice or likeness, or the copyrighted recordings used to create the training clips. See Rights and Responsible Use before using or sharing outputs.

Intended Use

For non-commercial research, education, and clearly labeled fan demonstrations, provided the user has the rights and permissions needed for their use. Generated audio should be disclosed as synthetic and unofficial.

Out-of-Scope Use

• Commercial use without all necessary permissions.
• Deceptive impersonation, fraud, or falsely attributing generated speech to a real person or rights holder.
• Suggesting that the model or its outputs are official, endorsed, or performed by Peter Cullen or the Transformers rights holders.
• Uses that violate applicable law, platform rules, or the rights of others.

Rights and Responsible Use

The training clips were selected from dialogue in Transformers films and television, including Bayverse films and Transformers: Prime. Those recordings are copyrighted by their respective rights holders. The audio clips and training dataset are not included in this repository. The model's CC BY-NC 4.0 label is not a representation that the uploader owns or has cleared rights to the source recordings, character, trademarks, or voice likeness. Users are responsible for obtaining any permissions required for their intended use; the model may be subject to takedown or other rights claims.

Limitations

• The model synthesizes English (en-us) only and may mispronounce names, unusual spellings, or words outside its training domain.
• Prosody, emotion, and pronunciation may differ from the character or source performer; outputs are not recordings of the performer.
• Evaluation was limited to project-level listening comparisons against the base voice. No standardized benchmark, independent evaluation, or quantitative quality metric is reported.
• The model may reproduce biases or artifacts present in the base voice and training material.

Audio Sample

Generated with the released best-loss checkpoint using this text:

Freedom is the right of all sentient beings.

Download the WAV sample.

Usage

Download optimus-final.onnx and its companion optimus-final.onnx.json, install a compatible Piper runtime, then synthesize audio with:

printf '%s\n' 'Freedom is the right of all sentient beings.' | \
  piper --model optimus-final.onnx \
        --config optimus-final.onnx.json \
        --output_file output.wav

The output is 22,050 Hz mono audio. Check the Piper runtime's documentation for installation and version-specific options.

Training

The model was fine-tuned from the Lessac medium checkpoint using Piper's VITS training pipeline. The project configuration records a learning rate of 1e-5, batch size 4, CPU training, eSpeak en-us phonemization, and a 22,050 Hz sample rate. Training audio was sliced into clips, transcribed, and manually reviewed for text and unwanted music, sound effects, or non-target speech. Exact dataset size, training duration, and quantitative evaluation results are not reported. The audio and dataset metadata are not distributed with the model.

Source

• Project repository
• Base voice