# Voice Changer
> Replace the voice in an audio recording — or in a whole talking video — with a different voice, preserving the original emotion, cadence, and timing.

## Overview

The Voice Changer node uses ElevenLabs Speech-to-Speech to re-voice media. It comes in two modes, chosen automatically by what you wire into it:

- **Audio mode** — wire **audio** in, get re-voiced **audio** out. The classic speech-to-speech path.
- **Video mode** — wire a **video** in, get the **video back with a new voice** (plus the new audio track on a separate handle). The node demuxes the audio from the clip, runs speech-to-speech, and remuxes the new voice onto the original video for you — no separate extract-audio / merge-video nodes needed.

In both cases the target voice's identity is applied while the original pacing, intonation, and emotional delivery are preserved.

## Configuration

| Field | Type | Default | Description |
|-------|------|---------|-------------|
| Voice | `string` | `""` | Target voice to apply. Selectable via VoiceBrowser (premade, custom, or library) |
| Voice Type | `"premade" \| "custom" \| "library"` | `"premade"` | Source of the selected target voice |
| Model | `"eleven_english_sts_v2" \| "eleven_multilingual_sts_v2"` | `"eleven_multilingual_sts_v2"` | ElevenLabs speech-to-speech model. **Multilingual v2** covers 29 languages and is ElevenLabs' recommended model — including for English source audio, where it often outperforms the English-only model. **English v2** remains selectable. Credit cost is the same (4 credits) for either model. If the field is left unset, the backend falls back to `eleven_multilingual_sts_v2`. |
| Stability | `number` (0-1) | `0.5` | Voice consistency. Lower = more expressive, higher = more uniform |
| Similarity Boost | `number` (0-1) | `0.75` | How closely the output matches the target voice timbre |
| Style Exaggeration | `number` (0-1) | `0` | Amplifies the source speaker's stylistic delivery in the target voice. Keep at `0` unless you want extra drama — values above 0 add latency and can reduce stability. Only sent to the model when above its default. |
| Remove Background Noise | `boolean` | `false` | **Off** keeps the music / SFX bed under the new voice. **On** removes background and yields a clean, voice-only result. |

## Inputs & Outputs

- **Inputs:**
  - `audio` — source audio whose voice will be replaced (audio mode).
  - `video` — source video to revoice (video mode). **When both `audio` and `video` are wired, video wins and the audio input is ignored.**
- **Outputs:**
  - `audio` — the re-voiced audio track (always produced; in video mode this is the new dialogue track).
  - `video` — the re-voiced video. **Disabled until a video input is wired** (audio mode produces no video).

## Video Mode

Wire any talking video (a generated clip, an uploaded file, a lip-synced shot) into the **video** input and the node will:

1. Extract the audio track from the clip.
2. Re-voice it with your selected voice (speech-to-speech).
3. Remux the new voice onto the original video and return it — plus the new audio track on the `audio` output handle.

This collapses what used to be a four-node chain (generate video → extract audio → voice changer → merge audio+video) into a single node and a single result.

**Requires an audio track.** Most text-to-video / image-to-video models output *silent* video — only models that generate sound (e.g. Veo 3, Kling) produce a usable track. If you feed in a silent clip, the node fails fast with: *"This video has no audio track to revoice."* Use a clip with spoken audio, or feed audio directly.

**Keeping the music bed.** Leave **Remove Background Noise** off to keep any music or sound effects baked into the clip's audio under the new voice. Turn it on for a clean, voice-only result.

## Best Practices

- Leave **Model** on Multilingual v2 (the default) — ElevenLabs recommends it even for English source audio, and it's required for non-English audio. Switch to English v2 only if you want to compare results; both cost the same.
- Start with Stability at 0.5 and Similarity Boost at 0.75, then adjust based on results. Higher similarity produces more accurate voice matching but can reduce naturalness.
- Leave Style Exaggeration at 0 for most work — it carries the source delivery into the new voice but trades off latency and stability when pushed up. Nudge it up only when you want the new voice to lean harder into the original performance's drama.
- For video mode, prefer source clips whose audio is mostly speech. Heavy music under the dialogue can bleed into the conversion — turn on Remove Background Noise if you want a clean voice and don't need the original bed.
- Use clean, well-recorded source audio for best results. The model preserves delivery characteristics from the input, so poor input quality carries through.
- For dramatic voice changes (e.g., male to female), allow for some loss of nuance -- the further the source and target voices differ, the more processing artifacts may appear.

## Common Use Cases

- Re-dubbing a talking-head or lip-synced video with a different voice in one step
- Re-voicing content to match a brand voice or character
- Anonymizing speakers in recordings while keeping natural delivery
- Converting rough scratch tracks to polished voiceover
- Applying a consistent narrator voice across recordings from different speakers

## Tips

- The emotion and pacing of the original are preserved in the output -- this is Speech-to-Speech, not Text-to-Speech. The input performance matters.
- Custom cloned voices (created via the Voice Clone node) can be used as the target voice for personalized re-voicing.
- If the output sounds robotic or unnatural, try lowering the Similarity Boost to give the model more freedom.
- In video mode, both the revoiced **video** and the revoiced **audio** are available as outputs — wire whichever the rest of your workflow needs.
- This node works with any media input -- it does not need to come from another Nodaro node. Uploaded clips and externally hosted URLs both work.