> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nijta.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Voice Harbor

> Learn how to redact biometric voiceprints in audio file using Voice Harbor's API.

<CardGroup cols={1}>
  <Card title="Access Token" icon="key" href="https://voiceharbor.ai">
    Automatically redact biometric identifiers from audio for privacy compliance.
  </Card>
</CardGroup>

## Voice Harbor V4.0 Release

<Update label="September 2025 Release" description="Voice Harbor V4.0">
  | Category               | V3.0                  | V4.0                            |
  | ---------------------- | --------------------- | ------------------------------- |
  | Expressiveness         | Neutral only          | Emotional expressiveness        |
  | Language Support       | Core set              | Extended European languages     |
  | Control Parameters     | Age, gender, language | +Expressiveness control         |
  | Audio Quality          | 16 kHz                | 22 kHz (Mini + Advanced)        |
  | Word Error Rate (WER)  | Baseline              | −58% (Mini), −62% (Advanced)    |
  | MOS (Naturalness)      | 3.72                  | 4.15 (≈ +12%)                   |
  | Processing Speed (RTF) | \~1.5×                | Mini: \<0.5×, Advanced: \~0.75× |

  With **V4**, Voice Harbor delivers clearer, faster, and more natural anonymized speech—without sacrificing privacy. Both **Mini** (optimized for speed) and **Advanced** (optimized for quality) modes are now more efficient than ever, and both output at **22 kHz**.

  #### Features

  * **Biometric Voiceprint Redaction:** Stronger anonymization to protect identity.
  * **Prosody & Emotion Preservation:** Natural intonation with emotional expressiveness.
  * **Enhanced Multilingual Support:** Broader coverage across European languages.
  * **Faster & Higher Quality:** 22 kHz output with up to **3×** faster processing.

  #### VPC2024 Evaluation Scores

  | Model                 | **EER ↓** | **UAR ↑** | **WER ↓** |
  | --------------------- | --------- | --------- | --------- |
  | B1                    | 10        | 42        | 2.9       |
  | B2                    | 13        | 55        | 10.0      |
  | B5                    | 36        | 38        | 4.4       |
  | B6                    | 25        | 36        | 9.1       |
  | SHIVA Orange          | 41        | 41        | 7.8       |
  | South Korea           | 44        | 28        | 5.0       |
  | FreeVC                | 35        | 44        | 4.5       |
  | NTU-NPU               | 42        | 40        | 4.5       |
  | Voice Harbor V3.0     | 44        | 40        | 3.8       |
  | **Voice Harbor V4.0** | **42**    | **38**    | **2.7**   |
</Update>

<Update label="Mai 2025 Release" description="Voice Harbor V3.0">
  ## Changelog

  #### Features

  * **Biometric voiceprints Redaction:** Remove unique vocal traits to protect your identity.
  * **Emotion Preservation:** Retain the natural emotional tone of your speech.
  * **High-quality Voices:** Generate clear, natural-sounding audio.
  * **Generative Content Anonymization:** Edit words and phrases for a tailored message.
</Update>

<Update label="2024-04-01" description="Voice Harbor V2.0">
  ## Changelog

  [Voice Harbor V2.0](https://app.nijta.com/) introduces a cutting-edge approach to secure speech data transformation, prioritizing privacy and compliance. This version features advanced biometric identity redaction, ensuring sensitive audio content is anonymized while maintaining clarity and usability. Fully approved by CNIL, V2.0 upholds the highest standards for data protection, making it an ideal solution for a wide range of applications requiring secure voice processing.

  #### Features

  * **Biometric voiceprints Redaction:** Remove unique vocal traits to protect your identity.
  * **Regulatory Compliance:** Ensure privacy standards with CNIL-guaranteed protection.
</Update>