michi-ai 0.1.0__tar.gz → 0.1.1__tar.gz

{michi_ai-0.1.0 → michi_ai-0.1.1}/PKG-INFO

@@ -1,12 +1,12 @@
 Metadata-Version: 2.1
 Name: michi-ai
-Version: 0.1.0
+Version: 0.1.1
 Summary: Full-duplex speech LLM client for MichiAI
 Home-page: https://ketsuilabs.io
 License: Apache-2.0
 Keywords: llm,speech,full-duplex,audio
 Author: Damian Krystkiewicz
-Author-email: damian@ketsuilabs.io
+Author-email: 45499236+dkrystki@users.noreply.github.com
 Requires-Python: >=3.9,<4.0
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Programming Language :: Python :: 3
@@ -40,7 +40,7 @@ Unlike traditional serial pipelines (ASR → LLM → TTS), MichiAI can listen an
 | **Latency (TTFA)** | ~75ms (tested on RTX 4090) |
 | **Architecture** | Continuous Embeddings + Rectified Flow Matching |
 | **Base Backbone** | SmolLM-360m |
-| **Key Innovation** | No Coherence Loss / Single Forward Pass per Word |
+| **Key Innovation** | No Coherence Loss / Single Step Decoding |
 
 
 ## 🌟 Key Features
@@ -56,7 +56,7 @@ Unlike traditional serial pipelines (ASR → LLM → TTS), MichiAI can listen an
 ## 🤖 Architecture Overview
 
 ### 1. The Listening Head
-A multi-modal encoder mapping raw audio into a continuous embeddings while simultaneously generating text tokens. This ensures the model understands both the semantic meaning and the emotional context.
+A multi-modal encoder mapping raw audio into continuous embeddings while simultaneously generating text tokens. This ensures the model understands both the semantic meaning and the emotional context.
 
 ### 2. The Speaking Head
 Predicts audio embeddings using **Rectified Flow Matching**. This allows for fast, high-quality, and diverse speech generation. The embeddings are then processed through a lightweight, causal **HiFi-GAN vocoder** for real-time streaming.

{michi_ai-0.1.0 → michi_ai-0.1.1}/README.md

@@ -21,7 +21,7 @@ Unlike traditional serial pipelines (ASR → LLM → TTS), MichiAI can listen an
 | **Latency (TTFA)** | ~75ms (tested on RTX 4090) |
 | **Architecture** | Continuous Embeddings + Rectified Flow Matching |
 | **Base Backbone** | SmolLM-360m |
-| **Key Innovation** | No Coherence Loss / Single Forward Pass per Word |
+| **Key Innovation** | No Coherence Loss / Single Step Decoding |
 
 
 ## 🌟 Key Features
@@ -37,7 +37,7 @@ Unlike traditional serial pipelines (ASR → LLM → TTS), MichiAI can listen an
 ## 🤖 Architecture Overview
 
 ### 1. The Listening Head
-A multi-modal encoder mapping raw audio into a continuous embeddings while simultaneously generating text tokens. This ensures the model understands both the semantic meaning and the emotional context.
+A multi-modal encoder mapping raw audio into continuous embeddings while simultaneously generating text tokens. This ensures the model understands both the semantic meaning and the emotional context.
 
 ### 2. The Speaking Head
 Predicts audio embeddings using **Rectified Flow Matching**. This allows for fast, high-quality, and diverse speech generation. The embeddings are then processed through a lightweight, causal **HiFi-GAN vocoder** for real-time streaming.

{michi_ai-0.1.0 → michi_ai-0.1.1}/pyproject.toml

@@ -1,9 +1,9 @@
 [tool.poetry]
 name = "michi-ai"
-version = "0.1.0"
+version = "0.1.1"
 description = "Full-duplex speech LLM client for MichiAI"
 license = "Apache-2.0"
-authors = ["Damian Krystkiewicz <damian@ketsuilabs.io>"]
+authors = ["Damian Krystkiewicz <45499236+dkrystki@users.noreply.github.com>"]
 homepage = "https://ketsuilabs.io"
 repository = "https://github.com/KetsuiLabs/MichiAI"
 readme = "README.md"