Replica Preview 0.7B EN

Replica Preview 0.7B EN

Replica Preview 0.7B EN is an English text-to-speech (TTS) model from TokenAI, designed for natural speech generation and reference-based voice cloning.

Replica Preview 0.7B EN is part of the Replica Preview model family by TokenAI. Learn more about the family: https://tokenai.llc/projects/replica-preview

Model Details

Property Value
Model type English Text-to-Speech (TTS) with reference-based voice cloning
Architecture Autoregressive T3 + S3Gen
Parameters 746,425,718 tensor elements (~0.7B)
Primary language English
Output sample rate 24 kHz
Voice inventory No fixed speaker catalog; speaker identity is conditioned by a reference recording
Model ID replica-preview-0.7b-en
Ownership TokenAI

The complete active inference path totals 746,425,718 parameters (~0.74B), reported publicly as 0.7B.

Capabilities

  • English text-to-speech generation.
  • Reference-based voice cloning with a consenting speaker reference.
  • Natural conversational speech generation.
  • Support for short and long English utterances.
  • Reproducible generation with configurable temperature and sampling settings.

Voice Cloning

Provide a clean reference recording from one consenting speaker, preferably longer than five seconds and with limited background noise. The reference conditions the generated speaker identity; it does not create a fixed speaker catalog.

Voice similarity can vary with reference duration, recording quality, prosody, and text. Do not use generated speech to impersonate a person without authorization.

Recommended Settings

temperature=0.8
top_p=0.95
top_k=1000
repetition_penalty=1.2

These settings are the validated starting point for the current release. Adjust them only after checking intelligibility and speaker identity.

Validation

The current release was evaluated with regression comparisons, English voice-cloning stress samples, CPU/GPU inference measurements, clipping checks, and clean-directory loading.

The evaluation did not establish a validated speaker-similarity score, so no unsupported similarity claim is made here.

Audio Samples

Audio samples are not hosted or displayed in this repository. To hear the official audio samples β€” and to see the full technical details and all features of this model β€” visit the official Replica Preview 0.7B EN page on TokenAI's website: https://tokenai.llc/projects/models/replica-preview-0.7b-en

Limitations

  • Validated language scope is English; multilingual support is not claimed.
  • Gender, age, and accent coverage were not verified as labeled evaluation dimensions.
  • Speaker similarity varies with reference quality and duration.
  • Pause quality, repetition, and hallucination behavior require additional human or validated automated review.
  • The model may require GPU inference for low-latency production use.

Ownership

Replica Preview 0.7B EN is owned and developed by TokenAI as part of the Replica Preview model family.

Project page: https://tokenai.llc/projects/replica-preview

About TokenAI

TokenAI is a non-profit AI startup founded on May 10, 2025 by Assem Sabry. TokenAI builds advanced AI models, research, datasets, and developer tools. We focus on language, multimodal, speech, cybersecurity, and Arabic-native intelligence. Our mission is to make advanced intelligence accessible and useful for everyone.

Website: https://assem.one

Contact TokenAI

For inquiries, partnerships, or support, contact TokenAI at info@tokenai.llc.

Citation

@misc{replica_preview_07b_en_2026,
  title={Replica Preview 0.7B EN},
  author={TokenAI},
  year={2026},
  url={https://huggingface.co/tokenaii/Replica-Preview-0.7B-EN}
}

Architecture: autoregressive T3 + S3Gen.

Downloads last month
-
Safetensors
Model size
0.5B params
Tensor type
F32
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Collection including tokenaii/Replica-Preview-0.7B-EN