January 21st.MiniMax announced yesterday that it has brought the newly upgraded T2A-01 SeriesVoice Modeland globally synchronized launch of Conch Voice products.
According to the report, relying on the T2A-01 series of speech models, the user isConch AI Generate natural, smooth hyper-anthropomorphic vocals by simply typing text.Up to 10,000 charactersThe user is able to use the output voice to meet the needs of complex scenarios. At the same time, users can freely configure the mood, speech rate, pitch, and even adjust the tone effect of the output voice according to their needs, to meet the refined needs of complex scenes.
1AI notes that Conch Voice supports theChinese, Cantonese, English, Japanese, Korean, Arabic, Portuguese, Spanish, French, Italian, Indonesian, Dutch, Russian, German, Vietnamese, Turkish, Ukrainian and 17 other languages.generate.
Officially, the same evaluation set and evaluation tools as the Seed-TTS paper (WER: text and asr recognition results of synthesized audio to calculate the word error rate; SIM: speaker identification tool to calculate the distance between synthesized and real audio, indicating the replica similarity) were used to calculate Conch Speech's word error rate and similarity (the bolded numerical scores represent better results). The results show that.Conch voice has the best character error rate and similarity in Chinese, and the character error rate and similarity in English are close to the real recording "Human".
Conch Voice has the ability to understand emotions, intelligently recognizing and reproducing the subtle emotional differences in speech, making the voice more vivid and the expression of emotions more accurate. Meanwhile, 300+ tones in different languages are pre-programmed for users to choose from, and users can filter by language, accent, gender and age. The tones are versatile, with a wide variety of styles such as uninhibited, witty, and kind, and can be used in scenarios such as audiobooks, ASMR whispers, and news broadcasts.