# Changelog of Voice Match — speaker similarity & TTS consistency (`synthetic.ia/voice-match`) Actor

- **URL**: https://apify.com/synthetic.ia/voice-match/changelog.md
- **Full Actor documentation**: https://apify.com/synthetic.ia/voice-match.md

## Changelog

### 0.1.7 — 2026-09-17

- Added optional **AI-voice / deepfake detection** (wav2vec2 `MelodyMachine/Deepfake-audio-detection-V2`, CPU, baked into the image).
  - `detectAi` flag (default on): every compared audio also gets an `aiGenerated: { likelyAiGenerated, aiConfidence }` field, alongside the (primary) similarity result.
  - New `operation: "detect"` mode: check audios for AI generation with no reference needed (`audios[]`).
  - Warmed at startup in Standby; detection failure is isolated and never breaks the voice comparison.

### 0.1 — 2026-09-17

- First release. Speaker-similarity / voice-consistency checker (ECAPA-TDNN, CPU, no key, no GPU).
- Reference vs one-or-many candidates → cosine-similarity score + same/different-voice decision, ranked, best match flagged.
- Strictness levels (strict / tts / balanced / tolerant) + custom threshold. ffmpeg preprocessing (16 kHz mono, silence-trim, loudnorm).
- SSRF-guarded URL input + base64; per-candidate error isolation; pay-per-event per comparison, limit checked first.