Model Introduction
VAL v2
Upgrade note: Voices created with VAL v1 have been automatically upgraded to VAL v2 — no action needed. If the first generation with an upgraded voice takes longer than usual, the system is usually finishing the upgrade for that voice; please wait a moment.
Overview
VAL v2 is our latest-generation voice synthesis model. Compared with VAL v1, it adds full support for 44.1kHz studio-grade high-quality output and further improves emotional expressiveness, stability, similarity, naturalness, and semantic understanding across all scenarios and languages.
What's New in VAL v2
🎧 44.1kHz Studio-Grade Audio
In high-quality mode, VAL v2 can output 44.1kHz studio-grade audio, with noticeably better clarity, overall quality, and vocal detail for demanding professional productions.
🧠 Stronger Core Capabilities
Across texts and voices in every scenario and language, VAL v2 delivers stronger emotional expressiveness, stability, similarity, naturalness, and semantic understanding than VAL v1.
Carried-Over Capabilities
VAL v2 keeps and strengthens everything the VAL series is known for:
- Languages and dialects: Chinese, Cantonese, English, Japanese, Korean, French, German, Spanish, and Portuguese, plus 30+ accent and dialect variants; a single voice can speak across languages while keeping its timbre, accent, and delivery
- Film-grade performance: Emotional range, similarity, stability, naturalness, and semantic understanding at film-production level — including shouting, rap, playful tones, and ASMR
- Emotion control: Control the emotion of each passage and blend multiple emotions by ratio — even a flat reading sample can be given rich emotion
- Sample fidelity: Beyond the voice itself, the model reproduces the sample's acoustic environment (sense of space, reverb, loudness) for output that sounds closer to the original
- Professional cloning and voice conversion: Samples from tens of seconds to hours, and conversion of any audio (including singing) into a target voice
Get Started
Ready to try VAL v2? Head to our Quick Start guide and start creating.
VAL v1
Previous generation. Voices created with VAL v1 have been automatically upgraded to VAL v2.
Release Date: December 2025
Overview
VAL v1 represents our flagship voice synthesis model, delivering industry-leading performance across professional voice cloning and timbre conversion scenarios. This advanced model demonstrates exceptional capabilities in emotional expressiveness, stability, similarity, naturalness, and semantic understanding across diverse languages and use cases.
Key Features
🎭 Professional Voice Cloning
Our deep learning architecture analyzes every nuance of your audio samples - from subtle intonations and pronunciation patterns to rhythm, prosody, and vocal habits. Whether you provide a brief 30-second clip or hours of audio material, VAL v1 delivers cloning results that are virtually indistinguishable from the original voice.
Capabilities:
- Accepts audio samples ranging from tens of seconds to multiple hours
- Captures intricate vocal characteristics including tone, rhythm, and pronunciation style
- Produces synthesis quality that matches professional dubbing standards
- Maintains consistency across different content types and lengths
🌍 Multilingual Support
VAL v1 provides comprehensive support for over 30 global languages and dialects, enabling seamless voice synthesis across linguistic boundaries.
Supported Languages Include:
- English (US, UK, Australian variants)
- Chinese (Mandarin, Cantonese)
- Spanish, French, German, Italian
- Japanese, Korean
- Portuguese, Russian, Arabic
- And many more regional languages and dialects
🎨 Advanced Timbre Conversion
Transform any audio's vocal characteristics to match your desired voice profile with precision control over tonal qualities. This feature enables single voice actors to portray multiple characters convincingly.
Applications:
- Multi-character voice acting with a single performer
- Precise control over vocal expression and delivery
- Seamless compatibility with all existing voice characters
- Song cover creation through vocal conversion
- Creative audio production and content adaptation
💫 Superior Performance Metrics
VAL v1 achieves cinema-grade performance across all key evaluation criteria:
- Emotional Range: Rich emotional expression with natural tonal variations
- Voice Similarity: Industry-leading accuracy in matching target voices
- Stability: Consistent quality across different content and contexts
- Naturalness: Human-like speech patterns and flow
- Semantic Understanding: Contextual awareness for appropriate delivery
Use Cases
- Content Creation: Podcasts, audiobooks, video narration
- Entertainment: Character voices, dubbing, voice acting
- Business: Corporate training, presentations, IVR systems
- Creative Projects: Music covers, audio drama, experimental art
- Accessibility: Text-to-speech for visually impaired users
Getting Started
Ready to experience VAL v1? Visit our Quick Start guide to begin creating professional-quality voice content today.
The VAL model series is continuously updated. Check back regularly for the latest improvements.