Inflect-v2: 3.9M and 9.3M parameter open-weight TTS models (huggingface.co)

Inflect-v2 has just launched two compact open-weight text-to-speech (TTS) models, with 3.9M and 9.3M parameters, respectively. These models offer the ability to generate speech at speeds exceeding real-time on CPU, making them exceptionally efficient despite their small size. Their performance is notably competitive with larger lightweight TTS solutions like KittenTTS, Piper, and Supertonic-3, thereby presenting a significant advancement for developers and researchers looking for high-quality, efficient speech synthesis options. The models support a range of frameworks including CPU, CUDA, PyTorch, and ONNX, and are distributed under the Apache 2.0 license, promoting accessibility and collaboration within the AI/ML community. The Inflect-v2 models enable users to experiment with TTS technology without the need for extensive computational resources, as highlighted in their demo applications. This release not only democratizes access to powerful TTS capabilities but also underscores the ongoing trend towards optimizing AI models for performance while minimizing resource usage.