Skip to main content

Upload a Recorded Voice

Goal: Use a real human voice recording instead of AI TTS for a character — for professional voice actors, the filmmaker's own voice, or temp dialogue recorded on a phone.

ACT3 AI supports Azure Neural TTS by default, but you can replace it entirely with a real recording for any character. This is useful when you have hired a voice actor and want their performance to drive the lipsync, when you want to use your own voice as a scratch track while the project develops, or when a character's voice has qualities (a specific accent, a distinctive quality, an emotional rawness) that no TTS voice in the library matches.

Steps​

  1. Open the Casting panel. In the Editor, click the Cast icon in the left sidebar. Click the character you want to assign a recorded voice to.

  2. Go to the Voice tab. Click the Voice tab in the Character Detail view.

  3. Click Upload Recording instead of Browse Voices. The upload dialog opens with two sub-options: Upload Full Session (all lines recorded in a single file or session) and Upload Per Line (separate audio files for each line of dialogue). For most use cases, Upload Full Session is the right choice — ACT3 AI will handle splitting the recording into individual lines.

  4. Upload the audio file. ACT3 AI accepts WAV and MP3 formats. WAV is preferred for quality. Click Browse or drag the file into the upload area. Files should be recorded at 44.1 kHz or 48 kHz, mono or stereo (the system will convert to mono for processing).

  5. Understand how ACT3 AI processes the uploaded audio. Once the file uploads, the system automatically: trims leading and trailing silence, normalizes the audio level to a consistent loudness target, and runs a voice activity detection pass to identify individual speech segments.

  6. Match recordings to script lines (timestamp import vs manual sync). If you recorded a full session, ACT3 AI attempts to auto-match speech segments to the character's dialogue lines in the script using a phonetic alignment algorithm. Review the auto-matches:

    • Green segments are matched with high confidence.
    • Yellow segments are matched but should be reviewed.
    • Red segments were not matched and need to be assigned manually by dragging them to the correct line.

  7. Assign unmatched segments manually. For any red segment, click it to play it, then drag it to the corresponding dialogue line in the list. This is quick for a short script; for a feature-length voice session, auto-matching handles the bulk of the work.

  8. Understand lipsync generation from uploaded audio. Lipsync in ACT3 AI is always derived from the audio waveform — whether that waveform comes from TTS or from an uploaded recording. Once the recording is matched to script lines, lipsync is generated exactly as it would be for a TTS voice. The AI analyzes the phoneme timing in your recording and drives mouth movement accordingly.

  9. Replace a recording later. To update the recording (for example, to replace a temp scratch track with the final voice actor session), return to the Voice tab and click Replace Recording. Upload the new file and re-run the matching step. Shots with already-generated lipsync will need to be regenerated to reflect the new audio.

Tips​

  • Record in a quiet space with a consistent ambient noise floor — background noise in a reference recording will make phoneme detection less accurate and degrade lipsync quality. A closet full of clothes is a surprisingly effective vocal booth.
  • If you are recording a scratch track on your phone, speak clearly and at a normal conversational pace. Rushed temp tracks with mumbled words are harder to auto-match to script lines.
  • Do not clip the audio during recording. Set levels so the loudest line peaks around -6 dBFS, not 0. Clipped audio cannot be restored in post-processing.

Next steps​