Posts

Qwen Researchers Release Qwen3-TTS: an Open Multilingual TTS Suite with Real-Time Latency and Fine-Grained Voice Control

By -
Image
Alibaba Cloud’s Qwen team has open-sourced Qwen3-TTS, a family of multilingual text-to-speech models that target three core tasks in one stack, voice clone, voice design, and high quality speech generation. https://ift.tt/dX9W2TS Model family and capabilities Qwen3-TTS uses a 12Hz speech tokenizer and 2 language model sizes, 0.6B and 1.7B, packaged into 3 main tasks. The open release exposes 5 models, Qwen3-TTS-12Hz-0.6B-Base and Qwen3-TTS-12Hz-1.7B-Base for voice cloning and generic TTS, Qwen3-TTS-12Hz-0.6B-CustomVoice and Qwen3-TTS-12Hz-1.7B-CustomVoice for promptable preset speakers, and Qwen3-TTS-12Hz-1.7B-VoiceDesign for free form voice creation from natural language descriptions, along with the Qwen3-TTS-Tokenizer-12Hz codec. All models support 10 languages, Chinese, English, Japanese, Korean, German, French, Russian, Portuguese, Spanish, and Italian. CustomVoice variants ship with 9 curated timbres, such as Vivian, a bright young Chinese female voice, Ryan, a dynamic Eng...
Post a Comment
Read more

Microsoft Releases VibeVoice-ASR: A Unified Speech-to-Text Model Designed to Handle 60-Minute Long-Form Audio in a Single Pass

By -
Image
Microsoft has released VibeVoice-ASR as part of the VibeVoice family of open source frontier voice AI models. VibeVoice-ASR is described as a unified speech-to-text model that can handle 60-minute long-form audio in a single pass and output structured transcriptions that encode Who, When, and What, with support for Customized Hotwords. VibeVoice sits in a single repository that hosts Text-to-Speech, real time TTS, and Automatic Speech Recognition models under an MIT license. VibeVoice uses continuous speech tokenizers that run at 7.5 Hz and a next-token diffusion framework where a Large Language Model reasons over text and dialogue and a diffusion head generates acoustic detail. This framework is mainly documented for TTS, but it defines the overall design context in which VibeVoice-ASR lives. https://ift.tt/r7MyPqX Long form ASR with a single global context Unlike conventional ASR (Automatic Speech Recognition) systems that first cut audio into short segments and then run dia...
Post a Comment
Read more

FlashLabs Researchers Release Chroma 1.0: A 4B Real Time Speech Dialogue Model With Personalized Voice Cloning

By -
Image
Chroma 1.0 is a real time speech to speech dialogue model that takes audio as input and returns audio as output while preserving the speaker identity across multi turn conversations. It is presented as the first open source end to end spoken dialogue system that combines low latency interaction with high fidelity personalized voice cloning from only a few seconds of reference audio. The model operates directly on discrete speech representations rather than on text transcripts. It targets the same use cases as commercial real time agents, but with a compact 4B parameter dialogue core and a design that treats speaker similarity as a primary objective, not as an auxiliary feature. Chroma achieves a reported 10.96% relative improvement in speaker similarity over a human baseline and reaches a Real Time Factor (RTF) of 0.43, so it can generate speech more than 2 times faster than playback. https://ift.tt/z5sGVXw From cascaded ASR LLM TTS end to end S2S Most production assistants ...
Post a Comment
Read more

Inworld AI Releases TTS-1.5 For Realtime, Production Grade Voice Agents

By -
Inworld AI has introduced Inworld TTS-1.5, an upgrade to its TTS-1 family that targets realtime voice agents with strict constraints on latency, quality, and cost. TTS-1.5 is described as the number top ranked text to speech system on Artificial Analysis and is designed to be more expressive and more stable than prior generations while remaining suitable for large scale consumer deployments. Realtime latency for interactive agents TTS-1.5 focuses on P90 time to first audio latency, which is a critical metric for user perceived responsiveness. For TTS-1.5 Max, P90 time to first audio is below 250 ms. For TTS-1.5 Mini, P90 time to first audio is below 130 ms. These values are about 4 times faster than the prior TTS generation according to Inworld. The TTS-1.5 stack supports streaming over WebSocket so synthesis and playback can start as soon as the first audio chunk is generated. In practice this keeps end to end interaction latency in the same range as typical realtime language mode...
Post a Comment
Read more

How AutoGluon Enables Modern AutoML Pipelines for Production-Grade Tabular Models with Ensembling and Distillation

By -
In this tutorial, we build a production-grade tabular machine learning pipeline using AutoGluon , taking a real-world mixed-type dataset from raw ingestion through to deployment-ready artifacts. We train high-quality stacked and bagged ensembles, evaluate performance with robust metrics, perform subgroup and feature-level analysis, and then optimize the model for real-time inference using refit-full and distillation. Throughout the workflow, we focus on practical decisions that balance accuracy, latency, and deployability. Check out the  FULL CODES here . Copy Code Copied Use a different Browser !pip -q install -U "autogluon==1.5.0" "scikit-learn>=1.3" "pandas>=2.0" "numpy>=1.24" import os, time, json, warnings warnings.filterwarnings("ignore") import numpy as np import pandas as pd from sklearn.model_selection import train_test_split from sklearn.metrics import roc_auc_score, log_loss, accuracy_score, classifi...
Post a Comment
Read more

Liquid AI Releases LFM2.5-1.2B-Thinking: a 1.2B Parameter Reasoning Model That Fits Under 1 GB On-Device

By -
Image
Liquid AI has released LFM2.5-1.2B-Thinking, a 1.2 billion parameter reasoning model that runs fully on device and fits in about 900 MB on a modern phone. What needed a data center 2 years ago can now run offline on consumer hardware, with a focus on structured reasoning traces, tool use, and math, rather than general chat. Position in the LFM2.5 family and core specs LFM2.5-1.2B-Thinking is part of the LFM2.5 family of Liquid Foundation Models, which extends the earlier LFM2 architecture with more pre-training and multi stage reinforcement learning for edge deployment. The model is text only and general purpose with the following configuration: 1.17B parameters, reported as a 1.2B class model 16 layers, with 10 double gated LIV convolution blocks and 6 GQA blocks Training budget of 28T tokens Context length of 32,768 tokens Vocabulary size of 65,536 8 languages, English, Arabic, Chinese, French, German, Japanese, Korean, Spanish Reasoning first behavior and thinking traces...
Post a Comment
Read more

A Coding Guide to Anemoi-Style Semi-Centralized Agentic Systems Using Peer-to-Peer Critic Loops in LangGraph

By -
Image
In this tutorial, we demonstrate how a semi-centralized Anemoi-style multi-agent system works by letting two peer agents negotiate directly without a manager or supervisor. We show how a Drafter and a Critic iteratively refine an output through peer-to-peer feedback, reducing coordination overhead while preserving quality. We implement this pattern end-to-end in Colab using LangGraph, focusing on clarity, control flow, and practical execution rather than abstract orchestration theory. Check out the  FULL CODES here . Copy Code Copied Use a different Browser !pip -q install -U langgraph langchain-openai langchain-core import os import json from getpass import getpass from typing import TypedDict from langchain_openai import ChatOpenAI from langgraph.graph import StateGraph, END if not os.environ.get("OPENAI_API_KEY"): os.environ["OPENAI_API_KEY"] = getpass("Enter OPENAI_API_KEY (hidden): ") MODEL = os.environ.get("OPENAI_MODEL...
Post a Comment
Read more