Twoshot Voice Audio

Audiobook Production

Create audiobooks with voice cloning from a reference:

1. Text — save as script organized by chapters/sections, or fetch attached script
2. Voice reference — extract vocals from user's track using stem separation, or use an uploaded voice recording
3. Generate — voice cloning TTS with extracted vocals as reference + script text
   • Generate first chapter first — this IS the voice quality preview and real content
   • If the voice sounds off, regenerate with adjusted settings before continuing
4. Ambient layer — combine narration with background audio (ambient sounds, gentle music) in a studio project
5. Fallback — if voice extraction fails or no vocals found, offer standard narrator voice

Translation with Voice Preservation

Translate spoken content while keeping the original speaker's voice:

1. Transcribe — extract text from the original audio
2. Translate — convert transcript to target language, save as note for user review/editing
3. Voice clone — original audio as voice reference + translated text generates speech in the original voice but in the new language
4. Verify — cloned voice matches original in accent, pitch, and pacing
5. Fallback — if voice cloning isn't available, standard TTS preserves the content but not the voice

Voice Conversion

Change the speaker's identity while preserving the content:

• With a trained voice model: apply directly to the user's audio
• With a reference recording: voice swap model uses reference + input audio
• From a description: design the voice first (create a synthetic voice sample), then use as reference for voice swap

Preparing voice references:

• Search library for clean voice samples, or import from YouTube (interviews for speech, isolated vocals for singing)
• Requirements: clean solo voice, no background music/noise, 10-30 seconds of audio
• If reference has background music, separate vocals first
• If audio has instrumental, separate stems first, convert vocal only, merge back in a studio project — voice conversion works best on isolated vocals

For recurring voice personas, collect reference samples into an element: clean takes, different pitches/emotions, speaking and singing if both needed.

Vocal Enhancement

Clean up and enhance vocal recordings:

1. Separate — extract vocals from the mix (stem separation)
2. Dereverb — remove room reverb/echo for a dry signal
3. Enhance — apply quality enhancement/upscaling for clarity
4. Each step cascades — progressively cleaner results
5. If the target vocal is already dominant, offer cleanup/enhancement directly instead of separation

Text-to-Speech

Generate speech from text:

• Multiple voice models with different characteristics (pitch, pace, tone, accent)
• Style control through prompt description
• Save scripts as notes for review before generating
• Generate first line as preview, iterate on voice before producing full content
• For long-form content, break into segments and generate in parallel