# Premove ITN

> Context-aware inverse text normalization for voice agents

- [Context-aware inverse text normalization for voice agents](https://docs.premove.dev/index.md): Premove ITN turns spoken ASR transcripts into structured written values with sentence context. Open-source Python package and open model weights.
- [Premove ITN benchmark](https://docs.premove.dev/benchmarks.md): Frozen 1,500-row synthetic ITN stress benchmark, including 400 voice-agent cases, comparison results, latency, methodology, and limitations.
- [Premove ITN documentation](https://docs.premove.dev/docs/index.md): Install, integrate, and deploy context-aware inverse text normalization for voice agents.
- [Getting started with Premove ITN](https://docs.premove.dev/docs/getting-started.md): Install Premove ITN and normalize your first voice-agent transcript in Python or the CLI.
- [Python API](https://docs.premove.dev/docs/python-api.md): Load Premove ITN and normalize English voice-agent transcripts in Python.
- [Command-line interface](https://docs.premove.dev/docs/cli.md): Normalize direct text and newline-delimited transcripts with the Premove ITN CLI.
- [Supported forms](https://docs.premove.dev/docs/supported-forms.md): Structured values and English spoken forms supported by Premove ITN.
- [How Premove ITN works](https://docs.premove.dev/docs/how-it-works.md): How deterministic candidates, contextual scoring, and exact decoding work together.
- [Deployment](https://docs.premove.dev/docs/deployment.md): Deploy Premove ITN with correct model lifecycle, device, and platform expectations.
- [Troubleshooting](https://docs.premove.dev/docs/troubleshooting.md): Resolve common Premove ITN model, device, input, and output issues.
- [What is inverse text normalization?](https://docs.premove.dev/docs/learn/what-is-inverse-text-normalization.md): Learn how ITN turns spoken-form ASR transcripts into structured written text.
- [Why voice agents need inverse text normalization](https://docs.premove.dev/docs/learn/itn-for-voice-agents.md): How ITN turns voice transcripts into reliable tool and API arguments.
- [Premove ITN architecture](https://docs.premove.dev/docs/internals/architecture.md): Trace how Rust candidate generation, contextual DeBERTa scoring, and exact decoding normalize English voice-agent transcripts.
- [Structured prediction](https://docs.premove.dev/docs/internals/structured-prediction.md): How Premove ITN trains and decodes over complete compatible edit paths.
- [Premove ITN model provenance](https://docs.premove.dev/docs/internals/model-provenance.md): Review the production DeBERTa checkpoint, training stages, data sources, and frozen evaluation boundary.
- [Premove ITN inference artifact](https://docs.premove.dev/docs/internals/inference-artifact.md): Verify the v0.1.0 Hugging Face model export, immutable revision, and release artifact boundaries.
- [Rust candidate coverage](https://docs.premove.dev/docs/internals/rust-candidate-coverage.md): Inspect the English spoken forms and deterministic candidate kinds supported by the Premove ITN Rust layer.
- [Premove ITN platform support](https://docs.premove.dev/docs/internals/platform-support.md): Check validated macOS and Linux release targets, Python versions, inference devices, and certification evidence.

## Optional

- [GitHub](https://github.com/premove-ai/premove-itn)
- [PyPI](https://pypi.org/project/premove-itn/)
- [Hugging Face](https://huggingface.co/premove-ai/premove-itn)