Alibaba's Tongyi Lab Open-Sources Fun-CineForge, Solves Multi-Speaker Dubbing Challenge
Traditional AI voice dubbing often falls short in high-stakes productions like films and animation, where capturing nuanced emotional peaks and perfectly synchronized lip movements is paramount. To tackle this core industry challenge, Tongyi Lab has officially launched and open-sourced the groundbreaking film-grade, multi-scenario multimodal dubbing model—Fun-CineForge .
Bridging the Audio-Visual Gap: A Four-Pillar Framework for Seamless Sync
Rather than relying on basic text-to-speech, Fun-CineForge is engineered to master four critical dimensions of professional dubbing:
Lip Sync: Ensures synthesized speech aligns with on-screen character mouth movements with exceptional precision.
Emotional Expression: Infuses the voice with authentic human-like emotion by analyzing facial cues and contextual instructions.
Voice Consistency: Maintains a stable, recognizable vocal identity for specific characters across complex multi-speaker dialogue scenes.
Time Alignment: Enables millisecond-accurate insertion of dialogue, even when the speaker is off-screen or partially obscured.
Core Innovation: Pioneering "Time Modality" and a High-Fidelity Dataset
The technical leap of Fun-CineForge stems from its unique "data + model" co-design philosophy:

The CineDub High-Quality Dataset: Tongyi Lab has also open-sourced the automated CineDub dataset construction pipeline. Utilizing a chain-of-thought error correction mechanism, it reduces transcription error rates for Chinese and English text to approximately 1% - 2% and slashes speaker diarization errors to as low as 1.2%.
Four-Modality Fusion Architecture: The model pioneers the integration of a "time modality", jointly modeling visual inputs (lip shape and expression), text (dialogue and emotional context), and audio (voice reference). This fusion allows for exact synchronization in challenging scenes, including those without visible faces.
Demonstrated Excellence: Pioneering Authentic Multi-Character Dialogue Dubbing
Benchmark results demonstrate that Fun-CineForge substantially outperforms baseline models like DeepDubber-V1 across key metrics: word error rate (WER/CER), lip synchronization (LSE-C/D), and voice similarity. A landmark achievement is its first-of-its-kind capability to handle duet and multi-person dialogues with precision, showing remarkable robustness in video clips up to 30 seconds.
GitHub: https://github.com/FunAudioLLM/FunCineForge
HuggingFace: https://huggingface.co/FunAudioLLM/Fun-CineForge
ModelScope: https://www.modelscope.cn/models/FunAudioLLM/Fun-CineForge/
Related article
Google Unveils Veo 3.1 Lite Video Mode, Granting Ultra Subscribers Unlimited Generation
Google has introduced a new video generation mode, "Veo 3.1 - Lite [Low Priority]," exclusively for Ultra subscribers to boost creative output and cost efficiency. This mode’s key advantage is that it operates without consuming additional subscriptio
ChatGPT Deep Integration Unlocks 300,000 Movies and TV Shows via Conversation
As streaming services saturate the market with endless content, helping users cut through the noise to find what they actually want has emerged as a critical industry hurdle.On April 9, major streaming providers officially announced the integration o
How to fix Core Web Vitals for better SEO ranking
Transform your video editing workflow with AI-powered transcriptsTable of Contents:IntroductionKey Advantages of Automated TranscriptionEnsuring Accuracy and Refining OutputLeveraging Search Capabilities in TranscriptsSeamless Import of Existing File
Related Special Topic Recommendations
Comments (0)
0/500
Traditional AI voice dubbing often falls short in high-stakes productions like films and animation, where capturing nuanced emotional peaks and perfectly synchronized lip movements is paramount. To tackle this core industry challenge, Tongyi Lab has officially launched and open-sourced the groundbreaking film-grade, multi-scenario multimodal dubbing model—
Bridging the Audio-Visual Gap: A Four-Pillar Framework for Seamless Sync
Rather than relying on basic text-to-speech, Fun-CineForge is engineered to master four critical dimensions of professional dubbing:
Lip Sync: Ensures synthesized speech aligns with on-screen character mouth movements with exceptional precision.
Emotional Expression: Infuses the voice with authentic human-like emotion by analyzing facial cues and contextual instructions.
Voice Consistency: Maintains a stable, recognizable vocal identity for specific characters across complex multi-speaker dialogue scenes.
Time Alignment: Enables millisecond-accurate insertion of dialogue, even when the speaker is off-screen or partially obscured.
Core Innovation: Pioneering "Time Modality" and a High-Fidelity Dataset
The technical leap of Fun-CineForge stems from its unique "data + model" co-design philosophy:

The CineDub High-Quality Dataset: Tongyi Lab has also open-sourced the automated CineDub dataset construction pipeline. Utilizing a chain-of-thought error correction mechanism, it reduces transcription error rates for Chinese and English text to approximately 1% - 2% and slashes speaker diarization errors to as low as 1.2%.
Four-Modality Fusion Architecture: The model pioneers the integration of a "time modality", jointly modeling visual inputs (lip shape and expression), text (dialogue and emotional context), and audio (voice reference). This fusion allows for exact synchronization in challenging scenes, including those without visible faces.
Demonstrated Excellence: Pioneering Authentic Multi-Character Dialogue Dubbing
Benchmark results demonstrate that Fun-CineForge substantially outperforms baseline models like DeepDubber-V1 across key metrics: word error rate (WER/CER), lip synchronization (LSE-C/D), and voice similarity. A landmark achievement is its first-of-its-kind capability to handle duet and multi-person dialogues with precision, showing remarkable robustness in video clips up to 30 seconds.
GitHub: https://github.com/FunAudioLLM/FunCineForge
HuggingFace: https://huggingface.co/FunAudioLLM/Fun-CineForge
ModelScope: https://www.modelscope.cn/models/FunAudioLLM/Fun-CineForge/
Google Unveils Veo 3.1 Lite Video Mode, Granting Ultra Subscribers Unlimited Generation
Google has introduced a new video generation mode, "Veo 3.1 - Lite [Low Priority]," exclusively for Ultra subscribers to boost creative output and cost efficiency. This mode’s key advantage is that it operates without consuming additional subscriptio
ChatGPT Deep Integration Unlocks 300,000 Movies and TV Shows via Conversation
As streaming services saturate the market with endless content, helping users cut through the noise to find what they actually want has emerged as a critical industry hurdle.On April 9, major streaming providers officially announced the integration o
How to fix Core Web Vitals for better SEO ranking
Transform your video editing workflow with AI-powered transcriptsTable of Contents:IntroductionKey Advantages of Automated TranscriptionEnsuring Accuracy and Refining OutputLeveraging Search Capabilities in TranscriptsSeamless Import of Existing File





Home






