Voice Craft

On October 20th, the Ant Ultimate announced today the Open Source Ming-3.0 Voice Series, which provides voice synthesis, audio understanding and dialogue, and voice recognition series of five models. Developers can choose between tasks or combine different competencies into their own work streams. The Voice Design of the Ming-3.0-ttts series can describe the target sound in text while also supporting the audio setting of the reference audio. More specifically, the creator can use the director's mode to describe the way the lines are performed, how emotions change and what words need to be read again..

On October 15th, SpaceXAI released Grok Voice 3.0 today, officially known as the world's most intelligent voice-to-mouth model. The model has a new record in the handling of passenger calls, with industry-leading delays of less than one second and operating costs lower than other front-line models. Grok Voice supports millions of conversations daily in Tesla and SpaceX and is now online through the application interface, pricing consistent with the previous generation of products. Independence in Artificial Analysis..

On September 15th, the StepAudio3 series of models, including StepAudio3realtime, StepAudio3ASR, StepAudio3TTS, StepAudio3Gen and StepAudio3Music, were officially launched today, with multiple models ranked first on the Artificial Analysis list. All five of the above models are now on top of the skyrocketing platform. Officially, StepAudio 3 series is separate..

On May 8th, OpenAI released three real-time voice models for voice reasoning, real-time translation and streaming: GPT-Realtime-2: building voice intelligence for the production environment. (b) GPT-Realtime-Translate: supports real-time translation in more than 70 input and 13 output languages to break language barriers and help people communicate more naturally

News from May 7, yesterday, Inworld AI launched a new generation of voice models Realtime TTS-2, which are open to developers through Inworld API and Inworld Realtime API in the form of research previews. The core change of TTS-2 is the shift from a one-way text to a closed-ringed real-time dialogue structure: the model directly receives the actual audio in the conversation, thereby understanding the tone, rhythm and emotional state of the user and adjusting it accordingly. New version with four new capabilities: Voice D..

On December 16, in a message, the Master Model was announced by the official public sign, two “hundred-hear” voice models were officially opened and two models were upgraded. According to the introduction, it takes three seconds to get your voice seamlessly transacted in languages, dialects and emotions -- Mandarin, Chinese, Japanese, English, happy, angry nine languages, 18 dialects. Upgrade Fun-CosyVoice3 Model Upgrade: First package delayed reduction of 50%, doubling the accuracy of Chinese and English and supporting 9 languages 18 dialects, translingual cloning and emotional control; ..

On September 19th, yesterday afternoon, the smart face wall announced a new series of "Small Steel Guns": the VoxCPM base model for voice generation of 0.5B parameter sizes was introduced. The VoxCPM base model for voice generation was officially launched at the Human Voice Interactive Laboratory of the International Graduate School of Shenzhen, Singhua University. The model parameter size is 0.5B, with an industry SOTA level in terms of voice naturality, sound similarity and rhythm performance. Performance performance: RTF ≈ 0.17, support current output VoxCPM in Seed-TTS-EV..

On Thursday, August 29, Microsoft's Artificial Intelligence division officially launched its first two homegrown AI models -- the MAI-Voice-1 voice model and the MAI-1-preview general-purpose model. According to Microsoft, the new MAI-Voice-1 voice model requires only a single GPU to generate a minute-long audio in less than a second, while the MAI-1-preview model "gives users a glimpse of Copilot's future functionality". Currently, Microsoft has made the MAI-Voice-...

March 21 news, OpenAI yesterday (March 20) released a blog post, announcing the launch of speech-to-text (speech-to-text) and text-to-speech (text-to-speech) models, to improve voice processing capabilities, support developers to build more accurate, customizable voice interaction system, and further promote the commercialization of AI voice technology applications. In terms of speech-to-text models, OpenAI has launched gpt-4o-transcribe and gpt-4o-mini-transcribe...

January 21st, MiniMax announced yesterday that it has brought the newly upgraded T2A-01 series of voice models and launched Conch AI products globally. According to the introduction, relying on the T2A-01 series of voice models, users can generate natural and smooth super humanoid voices by inputting text in Conch AI, and the maximum length of input can be up to 10,000 characters. At the same time, users can freely configure the mood, speech rate, pitch, and even adjust the timbre effect of the output voice to meet the refined needs of complex scenarios. 1AI notes that Conch Voice supports Chinese,...

Wisdom Spectrum announced the launch of GLM-4-Voice end-to-end emotional voice model. Officially, GLM-4-Voice is able to understand emotions, express and resonate emotions, self-adjust its speech rate, support multiple languages and dialects, have lower latency, and can be interrupted at any time, which can be experienced by users on the "Wisdom Spectrum Clear Speech" App from now on. According to the introduction, GLM-4-Voice has the following features: Emotional expression and emotional resonance: the voice has different emotions and subtle changes, such as happy, sad, angry, scared, etc. Adjusting speech speed: In the same round of conversation, you can ask TA to speak faster or slower...

Recently, Alibaba launched a new open source voice model Qwen2-Audio based on its Qwen-Audio. This model not only performs well in voice recognition, translation and audio analysis, but also achieves significant improvements in functions and performance. Qwen2-Audio provides a basic version and a command fine-tuning version. Users can ask questions to the audio model through voice, and recognize and analyze the content. For example, users can ask a woman to say a paragraph, and Qwen2-Audio can determine her age or analyze her emotions; if a noisy voice is input…

Recently, a voice model called VoiceCraft has attracted widespread attention in the industry. According to official announcements, the performance of this model has surpassed XTTS, which undoubtedly brings new breakthroughs in the field of AI audio processing. Project address: https://github.com/jasonppy/VoiceCraft The biggest highlight of VoiceCraft is its powerful audio cloning ability. Users only need to provide a piece of original audio, and VoiceCraft can use deep learning technology to copy new audio that is extremely similar to the original audio.