AI Narration Studio

Story Voice Studio

Use OpenAI for instruction-driven narration or Chatterbox for voice cloning from a reference sample. Your 24-word scene workflow stays in the same app.

Narration script

0 / 4096 characters 0 words 0 scenes

Voice instructions

Speak ONLY the exact narration text provided in the script. Never add, invent, repeat, paraphrase, introduce, or comment on any words that are not present in the script.

Read the script as a natural, compelling storyteller.

The script is divided into paragraphs of exactly 24 words each. Each paragraph represents one separate video scene.

Read each 24-word paragraph as one complete narration segment.

At the end of every paragraph, create a clearly noticeable silent scene break before beginning the next paragraph.

This scene-break pause must be substantially longer than a normal comma pause or sentence-ending period pause. It should sound and appear in the audio waveform as an obvious separation between two different narration segments.

Do not attempt to maintain an exact number of seconds. Instead, make every paragraph break long enough that it is immediately recognizable as a scene transition.

Normal punctuation within a paragraph should use ordinary natural pauses. Paragraph breaks should use a distinctly longer pause.

Read one complete 24-word paragraph naturally. Stop speaking completely. Leave a clearly noticeable silent scene-break gap. Begin the next 24-word paragraph. Repeat throughout the entire script.

Never announce, describe, or verbalize the pauses.

Do not produce audible breathing, inhaling, exhaling, mouth noises, lip smacks, or vocal preparation sounds during the scene breaks. The gap between paragraphs should contain clean silence.

Use natural, expressive storytelling with conversational pacing, varied emphasis, emotional engagement, and smooth delivery.

Create the feeling of human storytelling through pacing, tone, emphasis, and rhythm—not through audible breathing.

Do not combine paragraphs or rush from one paragraph into the next.

Avoid robotic cadence, repetitive rhythm, and announcer-style delivery.

Never read formatting, paragraph markers, timing instructions, or any text that is not part of the narration itself.

Engine

Model

gpt-4o-mini-tts

Voice

Create Custom Voice

Requires OpenAI custom-voice access. The consent recording and sample must be the same speaker. This app sends uploads through your authenticated server; your API key remains hidden.

English consent phrase Consent name

Consent recording

Record only the exact phrase above. Maximum 30 seconds. For this Vercel app, keep the file under 4 MB.

Custom voice name

Voice sample

Use 10–30 seconds of clean speech. OpenAI requires at least 5 seconds of actual speech. Keep the file under 4 MB in this app.

Speed

Audio format

Generated audio

Download audio

Voice Polisher

Chatterbox-aware polishing: resample to 48 kHz, preserve consonant transients, use light de-essing and compression, then add subtle harmonic air. Processing happens in your browser and downloads as WAV.

Optional audio file

Rumble Cut (Hz)

Body / Warmth (dB)

Mud Control (dB)

Crispness / Presence (dB)

Static Sibilance EQ (dB)

De-Esser Frequency (Hz)

De-Esser Threshold (dB)

De-Esser Ratio

Max S Reduction (dB)

Air / Shine (dB)

Harmonic Air (%)

Compression Threshold (dB)

Compression Ratio

Attack (ms)

Release (ms)

Polish Mix (%)

Final Peak (dBFS)

Chatterbox Crisp is tuned to keep the raw voice intact: light cleanup, very gentle S control, slow-attack compression, subtle generated upper harmonics, then peak normalization. Harmonic Air creates new high-frequency content after 48 kHz resampling instead of simply boosting frequencies that are missing.

The API key stays on Vercel as an environment variable and is never sent to the browser. Generated speech is AI-generated. Podcast polishing runs locally in your browser.