en.Wedoany.com Reported - On July 7, Xiaomi, a Chinese smart terminal and AI technology company, launched the MiMo-V2.5-ASR full-chain voice model auditory base. The domestic API pricing is 0.5 RMB per hour, while the overseas pricing is 0.074 USD per hour. The model has opened its API and supports usage in the Token Plan, targeting applications such as speech transcription, voice interaction, meeting recording, content processing, multilingual recognition, and complex acoustic scenarios.
The MiMo-V2.5-ASR is not positioned as a simple speech-to-text tool, but as an auditory base model within Xiaomi's MiMo open platform. As speech recognition enters the AI application chain, developers need to handle not only clear Mandarin recordings but also real-world scenarios including dialects, bilingual code-switching, background noise, multi-speaker conversations, far-field audio pickup, lyrics recognition, and specialized terminology recognition. By offering this model as an API service, Xiaomi enables developers to integrate voice capabilities into applications, terminals, office systems, content platforms, and smart devices, with billing based on audio transcription duration.
Complex scenario recognition is the main selling point of this launch. MiMo-V2.5-ASR covers Chinese dialects such as Cantonese, Wu, Southern Min, and Sichuanese, and can perform recognition in Chinese-English code-switching scenarios without requiring pre-set language selection. In song recognition scenarios, the model can process Chinese and English lyrics as well as mixed audio of accompaniment and vocals. Under strong noise and far-field audio pickup conditions, the model emphasizes robust recognition capabilities. In multi-speaker cross-talk scenarios, it achieves more accurate transcription. Knowledge-related content such as classical poetry, technical terms, personal names, and place names is also included within its recognition capabilities.
Native punctuation capability affects the actual user experience. Many speech recognition products can convert audio to text, but the output often requires manual punctuation, sentence segmentation, and formatting. MiMo-V2.5-ASR emphasizes direct punctuation output by combining prosody and semantics, making transcriptions closer to ready-to-use text. Scenarios such as meeting minutes, interview transcription, classroom notes, voice memos, podcast subtitles, and customer service quality inspection all require more complete sentence segmentation and punctuation processing capabilities; otherwise, subsequent editing costs increase significantly.
The domestic API pricing of 0.5 RMB per hour also makes this model more accessible for batch application testing. The cost of speech recognition services typically directly affects whether developers are willing to integrate it into high-frequency scenarios, especially long-form audio transcription, meeting recording processing, call quality inspection, course subtitle generation, and audio/video content processing. By billing based on audio transcription duration, enterprises and developers can control costs based on actual usage volume and more easily evaluate the model's input-output ratio across different business scenarios.
Voice models are becoming a critical foundational capability for smart terminals and AI applications. Smartphones, smart speakers, in-vehicle systems, wearable devices, office software, customer service systems, and content creation tools all require more stable auditory input. Xiaomi has previously deployed text, voice, and multimodal capabilities on the MiMo open platform. With the launch of MiMo-V2.5-ASR, the platform's voice input and understanding chain is further completed. With the API open, developers can build applications around scenarios such as dialect recognition, bilingual interaction, strong-noise transcription, and multi-speaker meeting recording.
After this model's launch, usage scenarios will initially focus on long-form audio transcription, mobile voice input, office meetings, content subtitles, customer service calls, and smart terminal voice interaction. Dialect coverage, Chinese-English code-switching, robust recognition under strong noise, and native punctuation output will directly determine the model's usability in real-world scenarios. Once MiMo-V2.5-ASR enters the open API phase, developers can conduct integration testing through the platform and use the relevant speech recognition capabilities based on transcription duration.
