Local Narrator TTS is a text-to-speech plugin for Adobe Premiere Pro by Shiki Tatsuta that generates AI narration locally on your computer and inserts the resulting WAV directly into the Premiere timeline. It uses a reference voice recording together with its exact transcript to generate new spoken lines that reflect the characteristics of the reference voice.

A key part of the workflow is local speech synthesis rather than sending every generation request to a cloud TTS service. Once the local engine and models are configured, editors can draft narration, hear it in context with their edit, revise the wording, and generate another take without repeatedly exporting and importing audio files.

FEATURES

• AI text-to-speech directly from Premiere Pro
• Local speech generation on your computer
• Qwen3-TTS as the default local engine
• Reference-voice-based narration generation
• Exact reference transcript controls
• Generates WAV audio files
• Inserts generated WAVs directly into the Premiere timeline
• Select the destination audio track
• Insert at the playhead
• Insert at a specified time in seconds
• Insert at the end of the sequence
• Preview narration in context and regenerate when needed
• Generated WAVs remain available for editing and reuse
• Custom output-folder support
• English and Japanese panel interfaces
• No cloud TTS per-generation fees

REFERENCE VOICE WORKFLOW

Local Narrator TTS does not simply require selecting a preset voice.

The workflow uses:

• A clean reference WAV
• An exact transcript of everything spoken in that WAV
• A matching spoken language
• The new narration text you want to generate

The developer recommends a reference recording of approximately six seconds or longer.

The reference and new narration should use the same language. The reference transcript must contain the actual words spoken in the selected recording rather than a summary or the new sentence you want to generate.

LANGUAGES

The Qwen language menu currently offers:

• English
• Japanese
• Chinese
• Korean
• French
• Spanish
• German

Use a matching-language reference recording and transcript for the narration language you intend to generate.

WORKFLOW

1. Install the plugin and local engine.
2. Open Local Narrator TTS in Premiere Pro.
3. Select a clean reference WAV.
4. Enter the exact transcript of the reference recording.
5. Select the matching spoken language.
6. Enter the new narration text.
7. Choose the destination audio track and insertion position.
8. Select Generate & Insert to Timeline.
9. The local engine generates a WAV and inserts it into the sequence.
10. Listen in context, revise the wording, or generate another take when necessary.

AUDIO STORAGE

By default, generated WAV files are stored in:

LocalNarratorTTS_Audio

This folder is created beside the saved Premiere Pro project. A different output directory can also be selected from Advanced Settings.

LOCAL ENGINE

Speech synthesis is processed on your computer. An internet connection is still required for initial setup, model downloads, and aescripts license activation or validation.

Local engine setup requires:

• Python 3.10–3.12
• FFmpeg available on PATH
• Several GB of free storage for the local engine, Python packages, and downloaded models

Once configured, speech generation runs locally without cloud TTS per-generation charges.

USES

• Draft voiceovers while editing
• Test narration before a final recording session
• Check whether dialogue fits an edit
• Generate narration for tutorials
• Experiment with alternative wording directly on the timeline
• Replace or revise individual lines
• Build rough voice tracks during editing
• Generate speech using your own voice or a voice you have permission to use

COMPATIBILITY

• Adobe Premiere Pro 25.6+
• Tested in Premiere Pro 2026
• macOS
• Windows
• Premiere UXP Panel
• Package includes the shared UXP .ccx and an Install/Quickstart PDF

NOTE

The current UXP release generates one clip at a time. Batch generation and multi-take variation are not included.

Pronunciation, pacing, output quality, and similarity to the reference can vary depending on the recording and individual generation. A clean reference, accurate transcript, and review of each generated clip are recommended before final delivery.

Use your own voice or a voice you have permission to use.