Voice Changer Pro

The Voice Changer Pro node gives every speaker in a conversation a new voice of their own. It detects each speaker in a recording or a talking video, recasts speaker 1 with your first voice, speaker 2 with your second voice, and so on. Each speaker keeps their original emotion, pacing and timing, so a whole dialogue scene, podcast or interview is re-cast in one run.

Voice Changer Pro runs on Nodaro Cloud. On a self-hosted install, the node shows a NODARO mark and runs through your Nodaro Cloud connection, billed to the connected nodaro.ai account.

When to use it

• Re-cast a dialogue video with several characters in one step.
• Replace every voice in a podcast or an interview with anonymous voices.
• Cast a scene with specific character voices, and keep the original performance.
• Turn a rough recording with several speakers into a polished voiceover.
• Recast the guests of a panel while the host keeps their own voice.

For a recording with one speaker, Voice Changer is enough.

Quick start

Inputs and outputs

When both inputs are wired, the video wins and the audio input is ignored.

The file decides the mode, not the input. Before any work starts, the node checks what the file really contains:

• An audio-only file on the Video input, such as a voice recording saved as .mp4 or .m4a, runs as audio. You get the re-voiced audio, the Video output stays empty, and nothing fails.
• A video on the Audio input has its sound extracted first and is treated as audio.

How speakers get their voices

The node finds each speaker in the recording and numbers them in the order they first talk. Your list of voices is matched to that order.

• Up to 8 voices. The list holds 1 to 8 entries, and at least one entry must be a real voice.
• Keep-slots. A Keep original entry leaves that speaker's voice unchanged, while later speakers are still recast.
• Shorter lists. Speakers past the end of the list keep their original voices.
• Edit the list. Move a voice with the up and down arrows, turn it into a keep-slot with Keep, or remove it.

If you are not sure who talks first, run the recording through Transcribe with Speaker Diarization on, and read the speaker labels.

Settings

Settings for each voice

Open Voice settings under a voice to tune that speaker alone. Leave a setting untouched to use the model's default.

How the background is kept

Before any voice is recast, the node always splits the recording into a voice track and a music-and-effects track. Preserve background music does not decide whether this split happens. It only decides whether the music track is mixed back under the new voices.

If music bleeds into the new voices, or a voice sounds thin, set Separation quality to Quality — finer separation. Keep Fast when speed matters or the voices already sound clean.

Add a room or an echo to the voices

Voice FX adds one effect to all the new voices together, before the music is mixed back in. The effect sits on the voices only, so with Preserve background music on, the music stays dry under a voice with reverb.

• Reverb spaces: Room (indoor), Bathroom, Car interior, Hall / Lobby, Concert Hall, Church, Cave, Arena / Stadium and Outdoor (open air). Wet / Dry mix from 0 to 100 sets how much room you hear.
• Character: Telephone, Megaphone / PA, Echo / Slap-back and Custom. Echo and Custom use Delay (ms), from 20 to 2,000, and Decay, from 0 to 1, where a higher decay gives more repeats.

The Audio FX page describes each preset.

Re-voice a whole video

Wire any talking video into the Video input. The node then:

1. Extracts the sound track from the clip.
2. Detects each speaker in the order they first talk.
3. Recasts each speaker with the matching voice.
4. Puts the new voices back on the original video, and returns the video and the new sound track.

The video needs a sound track. Most text-to-video and image-to-video models make silent video, and a silent clip fails at once. Use a clip with spoken audio, or wire a recording into the Audio input.

Credits

Only recast voices cost credits. Keep-slots and speakers past the end of your list pass through for free.

• A Recast voice is billed by the length of its stem. The stem runs from the start of the clip to that speaker's last line. The rate is 44 credits per minute, counted per second. Each voice is rounded up, and then the run's total, so a run can cost slightly more than the rate times its stem lengths.
• A Re-speak voice costs what the same text costs on Text to Speech with the same model. Every started block of 100 characters of the text it speaks costs 5 credits on Re-speak (v3) and 5 credits on Re-speak (v4). Each Re-speak voice is billed as at least 8 blocks.
• Minimum. Every Recast voice is billed as at least 6 seconds of speech, and so is every run.

You are charged for the stems that were actually converted and the characters that were actually spoken again, and never more than the amount held when the run started.

To quote a whole convert from code, ask for its own identifier instead of adding up per-voice prices, which can land a credit off. The identifier is voice-changer-pro:<slot>,<slot>,…, with one slot per recast voice, in any order:

• sts-<seconds> for a Recast voice, by its stem length, for example sts-26.76.
• respeak-<characters> for a Re-speak voice on v3.
• respeak-v4-<characters> for a Re-speak voice on v4, priced on v4's own rate.
• respeak-v4t-<characters> for a Re-speak voice on v4 with a language code, which speaks a translation. A translated voice reserves 1.5 times its characters, rounded up, because a line may be rendered twice. The server applies that itself, so send the speaker's transcript characters in every case.
• The bare sts, respeak, respeak-v4 or respeak-v4t when the measure is unknown. They are priced as one minute or as one block of 1,000 characters, as the reservation holds them.

POST /v1/credits/model-costs and GET /v1/credits/model-cost answer it with the credits the convert reserves, for example for voice-changer-pro:sts-26.76,respeak-v4-1200,respeak-v4t-800. More than 8 voices, a stem longer than an hour, too many characters on one voice, or a malformed slot lands in missing. Only v4 speaks a translation, so a respeak-v3t slot is malformed.

Use it as an app

Voice Changer Pro is also a standalone app at voice.nodaro.ai. Paste a link or upload a video, choose a voice for each speaker, and get the converted video back.

Tips

• Order the voices with care. The match is by position: the first voice goes to the first speaker to talk.
• Tune each speaker separately. Each voice has its own settings, so you can steady one speaker and keep another expressive.
• Balance loudness per speaker. Use Volume on each voice: Match source mirrors the original speaker, and Manual sets an exact level.
• Map fewer voices than speakers. Only the listed speakers change; everyone after them keeps their voice.
• Keep the performance. Recast converts speech to speech, so the original acting carries through. Use Re-speak (v3) or Re-speak (v4) only when you want a new performance, for example on an audio-only project.

Troubleshooting

One speaker sounds robotic. Lower that voice's Similarity or Stability, or choose a voice whose timbre is closer to the original speaker.

Music bleeds into the voices. Set Separation quality to Quality — finer separation.

The wrong speaker got a voice. The speakers are numbered by who talks first. Check the order with Transcribe and reorder the voices.

A self-hosted run is refused with nodaro_connection_required. The install has no Nodaro Cloud connection. Connect it with OAuth or an API key, as described in Cloud connect.

From the API

POST /v1/voice-changer-pro runs the node from code, on Nodaro Cloud. orderedVoices maps speaker N to entry N. An entry is a voice id, an object with per-voice settings, or null for a keep-slot. The settings object takes voiceId, engine ("sts", "v3" or "v4"), stability, similarityBoost, style, useSpeakerBoost, seed, volumeMode and volume. With engine: "v3", stability accepts only 0, 0.5 or 1, and similarityBoost, style and useSpeakerBoost are ignored. With engine: "v4", stability accepts any value from 0 to 1 and similarityBoost applies, while style and useSpeakerBoost are ignored. The result's sourceHasVideo tells you whether the run used video mode or audio mode.

On Nodaro Cloud, you can also run the same engine in three steps. You can then check the speakers before you pay for a recast, and mix the tracks before you render:

1. Analyze. POST /v1/voice-changer-pro/analyze separates the voices from the music and detects the speakers. It returns each speaker's time segments, word count and a transcript snippet, plus the detected language. It costs 11 credits.
2. Recast to stems. Call POST /v1/voice-changer-pro with output: "stems" and the analysis. The speakers are not detected again, so you can recast with other voices as often as you like. The result is one dry track per voice.
3. Export. POST /v1/voice-changer-pro/export renders the final video from your mix. Send up to 16 tracks, each with a gain from 0 to 200, muted, and a kind of voice or background, plus an optional voiceFx. At least one track must be unmuted. The video stream is copied, never re-encoded. Adjusting the mix is free; the export costs 2 credits.

The SDK methods are client.voices.analyze, client.voices.recast and client.voices.exportMix. The CLI commands are nodaro voice analyze, nodaro voice recast and nodaro voice export. The MCP tools are voice_changer_pro, voice_changer_pro_analyze and voice_changer_pro_export. See Voice and media.