The article examines how AI testing has shifted from simple functional checks to evaluating model reliability, fairness, robustness, and explainability, illustrating the shift with three detailed client cases—financial bias audit, automotive voice‑assistant stress testing, and medical‑imaging consistency verification.
The article evaluates Qianwen’s new desktop voice‑input system, showing how it filters filler words, understands screen context, executes AI commands, and generates structured text, PPTs, and Excel reports, positioning voice as a viable replacement for traditional keyboard typing.
The article explains that pretraining for speech models functions like a broad liberal‑arts education, teaching universal acoustic and linguistic patterns through next‑token prediction, joint audio‑text training, and mask‑or contrast objectives, while clarifying common misconceptions and highlighting data bias and the need for clean, task‑specific fine‑tuning.
The article argues that successful speech model training starts with understanding user scenarios, then selecting appropriate data, and finally choosing metrics, detailing six key questions, data sourcing strategies, evaluation criteria, and compliance considerations to avoid the misconception that sheer data volume guarantees performance.
The article introduces Microsoft’s open‑source VibeVoice project, detailing its long‑audio ASR‑7B and real‑time TTS‑0.5B models, the continuous speech tokenizer and next‑token diffusion techniques, and provides quick‑start instructions for online demos and local deployment via Hugging Face.
This guide walks through choosing the right voice‑agent architecture, implementing streaming ASR with WebSocket, triggering sentence‑by‑sentence TTS, wiring the three layers together via async generators, optimizing latency to under a second, and avoiding common pitfalls such as missing VAD and checkpoint persistence.
The article examines how AI‑powered voice input can replace keyboard typing in office settings, evaluates environmental constraints, compares two leading tools—Typeless and ShandianShuo—through feature lists, limitations, and real‑world usage scenarios, and concludes with practical advice on choosing the right solution.
The Conformer architecture blends global self‑attention with a depthwise separable convolution module in a Macaron‑style block, addressing the strong local time‑frequency structure and long sequence length of speech signals while keeping computational cost manageable for modern ASR systems.
This guide explains how to assemble NVIDIA's Nemotron Speech, RAG, and Safety models into a low‑latency, secure production AI agent stack, covering performance benchmarks, multimodal retrieval, safety data sets, integration code, and deployment options for cloud, on‑premise, and edge environments.
This article provides a detailed side‑by‑side analysis of the open‑source ASR tools FunASR and Qwen3‑ASR, covering team origins, model architectures, language coverage, speed, deployment requirements, and ideal use‑cases so readers can decide which solution fits their projects best.
The 2026 overview details lightweight, heavily quantized open‑source speech‑recognition models and CPU‑specific inference engines, offering concrete tips, model comparisons, and a concise selection guide that enable real‑time, GPU‑free ASR deployment with low latency and high stability.
A developer has re‑implemented the state‑of‑the‑art Qwen3‑ASR model in MLX, enabling native execution on Apple M1‑M4 chips with real‑time factors as low as 0.08, 4‑bit quantization speedups of 4.7×, multilingual support for 52 languages, and features such as word‑level timestamps and streaming transcription.
Microsoft’s newly open‑sourced VibeVoice‑ASR model can transcribe up to 60‑minute audio in a single pass, preserving global context while providing built‑in speaker diarization and timestamps, supports 50+ languages, offers custom hot‑word injection, and can be deployed via Docker, Gradio, or vLLM for high‑throughput API services.
This guide walks through configuring Alibaba Cloud credentials, implementing a FastAPI backend with email function calling, Alibaba OpenSearch, image generation via DashScope, speech recognition, and a responsive HTML/CSS/JavaScript front‑end that supports text chat, image recognition, image synthesis, and voice interaction.
This tutorial walks through setting up local speech recognition with OpenAI's Whisper and Vosk, leveraging Alibaba Cloud's ASR services, building a WebSocket server/client for real‑time audio streaming, capturing audio in the browser via MediaRecorder or RecordRTC, and performing speech synthesis with pyttsx3 and Alibaba's Sambert model.
The article explains how NVIDIA’s Nemotron Stack combines ultra‑fast speech recognition, multimodal retrieval, and advanced safety models into a unified, low‑latency pipeline, offering practical integration code, performance insights, and deployment options for turning experimental AI agents into production‑grade services.
GLM‑ASR‑Nano‑2512, a 1.5 B‑parameter open‑source speech‑recognition model released in December 2025, delivers state‑of‑the‑art accuracy on Chinese dialects and low‑volume audio, outperforms Whisper V3 on benchmark tests, runs on consumer GPUs, and provides detailed installation and deployment guides for transformers, vLLM and SGLang.
This guide explains the background of speech‑to‑text technology, introduces the large‑model streaming speech recognition API, walks through obtaining an API key, and provides detailed Go and Python code for establishing a WebSocket connection, sending full‑client and audio‑only requests, and parsing server responses.
Introducing Ming‑Flash‑Omni‑Preview, a 103‑billion‑parameter open‑source multimodal model built on a sparse MoE architecture that delivers state‑of‑the‑art performance in controllable image generation, streaming video understanding, and context‑aware speech recognition, surpassing prior models on GenEval and GEdit benchmarks.
This article introduces four high‑impact open‑source projects—a curated public dataset collection, the Kronos financial K‑line analysis model, WhisperLiveKit for real‑time speech transcription, and Youtu‑agent for building versatile AI agents—each with descriptions, key features, and GitHub links.
This tutorial walks through creating a Python program that captures spoken input, converts it to text with Baidu AI, sends the text to the free Qingyunke chatbot API for a response, and finally synthesizes the reply back into speech, complete with code snippets and setup instructions.
Learn how to create a Python program that captures spoken input, converts it to text using Baidu's speech‑recognition API, sends the text to the free Qingyunke chatbot for intelligent replies, and then synthesizes the response back into speech, with complete code snippets and setup instructions.
This tutorial walks you through creating a Python voice assistant powered by GPT‑4, covering project setup, virtual environment creation, required package installation, core code for speech recognition, text‑to‑speech, command handling, and optional speech‑rate adjustment.
This article explains how Baidu Map’s AI voice assistant converts spoken commands into precise navigation actions by detailing the speech‑to‑text pipeline, intent parsing, template and generative approaches, tool‑calling mechanisms, memory and reflection capabilities, and future directions for intelligent agents.
In a QECon 2025 talk, Ant Group’s AI team detailed a comprehensive multimodal evaluation framework that leverages large‑model metrics, custom pipelines, and benchmark datasets to assess image generation, speech recognition, and video quality, while also contributing to industry standards and academic research.
iQIYI’s AI research team secured three paper acceptances—two at ACL 2025 (including a main conference and a Findings paper) and one at INTERSPEECH 2025—covering long‑context large language model evaluation, Chinese novel summarization, and efficient Thai speech recognition, with links to each work.
This article provides a comprehensive guide on developing a Python tool to convert English videos into versions with Chinese dubbing and subtitles, covering all steps from audio extraction to final synthesis.
This article explains how modern AI advances have transformed audio processing, covering digital audio fundamentals, automatic speech recognition (ASR), text‑to‑speech (TTS), voice cloning techniques, and provides practical Python code examples using OpenAI Whisper and HuggingFace TTS models.
This article details the end‑to‑end development of an in‑vehicle voice assistant, covering motivation, functional design, technology stack selection, dialogue flow, privacy, third‑party integration, wake‑word detection, on‑device speech recognition, noise filtering, NLP processing, and deployment considerations.
Transformers, introduced in 2017, have reshaped neural networks by leveraging attention mechanisms to outperform RNNs and CNNs across NLP, computer vision, and speech tasks, offering parallel processing, long‑range dependency capture, and versatile applications such as translation, text generation, image classification, and speech recognition.
This article details HuoLaLa's development of an in‑house Automatic Speech Recognition system, covering its architecture, VAD optimization, language‑model and hot‑word enhancements, punctuation restoration, task and resource scheduling, and the resulting improvements in accuracy and cost efficiency.
This article introduces the Paraformer non‑autoregressive end‑to‑end speech recognition model released by Alibaba DAMO Academy, details its architecture, training strategies, large‑scale performance, and provides step‑by‑step guidance for using and fine‑tuning the model on the ModelScope platform with the FunASR toolkit.
This guide explains how to split a PDF into separate files using PyPDF2 and provides two Python approaches for converting MP3 audio to text—one leveraging Google Speech‑Recognition for higher accuracy and another using PocketSphinx for complete transcription—complete with ready‑to‑run code examples.
This article details the evolution from the initial Kaldi‑based speech recognition architecture (version 1.0) to a re‑engineered version 2.0, describing business background, service components, identified shortcomings, and a series of performance, concurrency, GPU, I/O, GC, and dispatch optimizations that dramatically improve resource utilization, latency, and reliability for large‑scale voice processing at 58.com.
This article explains the design, implementation, and experimental evaluation of hot‑word augmentation in WeNet's GPU runtime, detailing how character‑ and word‑based language model scoring are extended to boost recognition of rare proper nouns in both streaming and non‑streaming ASR services.
This guide explains how to integrate WeChat Mini Program voice capabilities by importing the recorder and audio APIs, recording audio, uploading for speech recognition, and playing back the result, with example code snippets for each step.
This article introduces the Paraformer model released by Alibaba DAMO Academy on ModelScope, detailing its non‑autoregressive architecture, training strategies, performance on benchmark datasets, and step‑by‑step guidance for fine‑tuning and deploying the model using FunASR and ModelScope pipelines.
The article chronicles the career of Chinese AI pioneer Huang Xuedong, detailing his education, rise at Microsoft, leadership of Azure AI, groundbreaking human‑level speech recognition breakthroughs, the engineering feats behind them—including a ten‑network model and the CNTK framework—and his recent move to Zoom.
The paper presents a low‑latency streaming speech‑recognition solution that reframes latency reduction as a knowledge‑distillation task, using a simple peak‑first regularization term to shift CTC output probabilities leftward and achieve up to 200 ms average latency reduction without harming word error rate.
Bilibili’s high‑quality ASR system combines large‑scale filtered business data, semi‑supervised Noisy‑Student training, an end‑to‑end CTC model with lattice‑free MMI decoding, and FP16‑optimized FasterTransformer inference on Triton, delivering top‑ranked accuracy, low latency, and scalable deployment for diverse Chinese‑English video content.
This article reviews the author’s six‑year hands‑on experience with deep learning, covering breakthroughs in speech recognition, computer vision, language modeling, reinforcement learning, privacy protection, model compression, recommendation systems, and future research directions, while summarizing technical lessons and practical insights.
This article surveys the most influential Transformer‑based research papers—from the original Attention Is All You Need work to recent models such as Autoformer and FEDformer—covering breakthroughs in natural language processing, computer vision, speech recognition, and long‑term series forecasting, and provides download links for each.
This article presents a comprehensive overview of the Efficient Conformer model for large‑scale end‑to‑end speech recognition, detailing its architectural improvements such as progressive downsampling and grouped multi‑head self‑attention, the PyTorch implementation in WeNet, streaming inference handling, experimental CER gains on AISHELL‑1 and production data, and future development plans.
The article details vivo's development of a high‑accuracy, high‑performance streaming speech‑recognition inference engine built on the wenet framework, its optimization techniques such as dynamic batching and memory pooling, collaborative acceleration with KunlunChip's high‑performance inference library, and extensive performance benchmarks demonstrating multi‑batch GPU and XPU gains.
58.com’s AI Lab presents a comprehensive overview of its end‑to‑end speech recognition system, detailing data collection, semi‑supervised training, Efficient Conformer architecture, model compression, and deployment strategies that together achieve high accuracy across diverse acoustic conditions and large‑scale production workloads.
In this interview, the chief algorithm expert at Zuoyebang discusses how AI technologies such as NLP, speech recognition, large‑model pre‑training, and knowledge‑graph construction are applied to K‑12 education, covering practical challenges, deployment strategies, and future research directions.
The AISummit AI conference highlighted advances in intelligent voice, with experts from ZuoYeBang, ByteDance, Microsoft and others discussing end‑to‑end speech recognition, pronunciation correction, and high‑quality speech synthesis, and exploring how multimodal pre‑trained models will shape the future of voice AI.
This report details a series of experiments on Chinese‑English mixed‑language speech recognition, introducing language‑identification loss and language‑model integration to improve acoustic modeling, reduce mixed error rates, and achieve significant gains over a baseline end‑to‑end ASR system.
This article presents a method to improve keyword detection in large‑scale speech recognition by integrating a prefix automaton into the beam‑search decoding of seq2seq models, enabling real‑time addition of new terms while reducing computational overhead compared to traditional approaches.
This article examines the challenges of English pronunciation teaching in China and presents a practical end-to-end phoneme‑level mispronunciation detection system that leverages CTC models, attention‑based text fusion, and data augmentation to dramatically reduce false alarms and improve diagnostic accuracy.
This tutorial explains how to create a Python voice chatbot by recording audio, converting speech to text with Baidu AI, sending the text to the Qingyunke chatbot API for a response, and finally synthesizing the reply back into speech using pyttsx3.
This article presents a detailed case study of using AI speech‑recognition techniques—including acoustic modeling with VGG, pypinyin conversion, feature extraction, and CTC decoding—to automatically verify game dialogue audio against script text, outlining the workflow, challenges, implementation details, and experimental results.
This talk presents a comprehensive study of invalid query detection in voice assistants, covering the definition of effective and ineffective queries, challenges of non‑human interaction and ambiguous intent recognition, data collection, model design, experimental results, user‑feedback loops, and future research directions.
This talk presents a comprehensive study of invalid query detection in voice assistants, covering the definition and taxonomy of invalid queries, challenges of non‑human interaction and ambiguous intent recognition, data collection and labeling strategies, feature engineering, deep neural network modeling, experimental results, user‑feedback loops, and current performance limits.
Baidu’s AI‑driven sign‑language digital human combines ultra‑accurate speech recognition, specialized translation, and precise gesture‑generation models—backed by extensive motion‑capture data and expert validation—to deliver 24‑hour, high‑fidelity signing for millions of hearing‑impaired users, showcasing inclusive AI communication.
This talk presents an overview of text‑plus‑speech multimodal emotion analysis, covering background, single‑modal text and audio models, the MSCNN‑SPU multimodal architecture, domain‑adaptation techniques, and future directions, with detailed model explanations, experimental results, and practical deployment insights.
This tutorial explains how to create a Siri‑style conversational robot in Python by configuring the environment, describing the speech‑recognition and chatbot principles, and showing the implementation that uses Baidu speech recognition and the Turing chatbot API.
KeSpeech, a benchmark jointly released by Beike AI and Tsinghua University at NeurIPS 2021, provides a massive Chinese Mandarin dialect dataset covering 30,000 speakers from 34 cities, supporting speech recognition, speaker verification, dialect identification, and voice conversion tasks, and includes rich multi‑scenario and parallel corpora for advanced research.
This article presents a comprehensive overview of real‑time voice dialogue systems, covering the hotline robot architecture, unique challenges of spoken interactions, ASR‑robust SLU models, multimodal emotion detection, oral expression handling, and the design and benefits of duplex (full‑duplex) conversational frameworks.
This article presents the business background of Tmall Genie’s voice‑driven content‑on‑demand service, critiques the traditional pipeline for entity extraction, and details an end‑to‑end speech‑semantic model—including the Speech2Slot architecture, knowledge‑enhanced encoding, and Phoneme‑BERT unsupervised pre‑training—demonstrating significant performance gains in both generation and classification tasks.
The article presents Tmall Genie’s end‑to‑end speech‑semantic understanding pipeline, detailing the limitations of traditional ASR‑NLU‑IR pipelines, introducing the Speech2Slot model with knowledge‑enhanced encoders, and describing unsupervised phoneme‑based pre‑training (Phoneme‑BERT) that improves entity extraction performance in voice‑driven content playback.
The article explains why traditional polling methods fall short for real‑time data, introduces the WebSocket protocol’s full‑duplex handshake and heartbeat mechanisms, and demonstrates how a Java‑based WebSocket service efficiently streams audio to an ASR engine for low‑latency speech recognition.
The article, authored by a former Stanford PhD now at Zoom, forecasts that by 2030 speech recognition will rely heavily on semi‑supervised learning, on‑device models, richer representations, and personalization, while applications such as transcription services and voice assistants will evolve modestly.
The article details the design, components, real‑time processing flow, and performance optimizations of 58.com’s streaming speech recognition engine, covering its SDK access layer, logical services, data storage, Kaldi‑based decoding, and the practical impact on voice‑driven applications.
This article describes the overall architecture of 58.com’s multi‑turn voice robot, explains rule‑based, intent‑based and text‑matching dialogue jump strategies, introduces an end‑to‑end classification approach using TextCNN, and reports its online performance improvements and future research directions.
The article summarizes two INTERSPEECH 2021 papers from Beike's voice technology team, detailing a grammar‑based semantic data augmentation method that improves end‑to‑end Chinese speech recognition and introducing GigaSpeech, a massive 10,000‑hour multilingual English speech dataset for robust ASR research.
This article presents the design, training, deployment, and evaluation of a self‑developed Voice Activity Detection system used in both real‑time streaming dialogues and offline audio analysis at 58.com, detailing algorithm choices, smoothing strategies, engineering challenges, and future research directions.
This article introduces the fundamentals of automatic speech recognition (ASR) for food‑sound classification, explains key audio representations and modeling approaches, and provides a fully runnable Python implementation using librosa, TensorFlow/Keras, and classic machine‑learning tools to train and predict on the Tianchi competition dataset.
The document details DiDi’s driver‑side intelligent voice assistant “XiaoDi,” describing its three‑layer architecture—audio source switching controller, semantic‑parsing core, and business API—along with conflict‑resolution mechanisms, multi‑turn dialogue handling, and a four‑region UI design that together enhance driver safety, convenience, and well‑being.
The article explores how modern AI technologies such as speech synthesis, natural language understanding, and the FastReID computer‑vision library enable realistic voice recreation and cross‑temporal dialogue, turning the emotional premise of the movie “Hi, Mom” into a tangible technical demonstration.
TAL Education (好未来) has opened a 587‑hour bilingual Chinese‑English speech dataset from classroom teaching, one of the largest open educational corpora, aiming to fill the data scarcity in mixed‑language speech recognition research and support AI model development.
This article details the AI-driven intelligent voice quality inspection system built by 58.com, covering its background, multi‑layer architecture, speech recognition, role and tag identification, backend services, and the resulting efficiency gains for large‑scale call‑center operations.
This article details how a three‑person team at 58.com built a self‑developed speech recognition engine in less than a year, covering background, team formation, data annotation, model selection, engineering architecture, performance optimizations, deployment results, and future directions.
This article describes the end‑to‑end pipeline of a voice robot, covering speech framing, feature extraction (FBank, MFCC), the VGGish embedding network, various model architectures, experimental results on accuracy and recall, and future directions for improving sound‑type classification.
This article explains the role of Weighted Finite State Transducers in conventional HMM‑based speech recognition, covering language models, pronunciation dictionaries, WFST definitions, semiring theory, composition and determinization operations, decoding graph construction (HCLG), lattice rescoring, and practical optimization techniques for real‑world scenarios.
This article explains the Kaldi chain model speech‑recognition system, covering HMM‑DNN fundamentals, discriminative (MMI) training, the special single‑state HMM topology, TDNN architecture, training pipelines, and experimental results that demonstrate its performance advantages over traditional GMM‑based approaches.
This article describes how Xiaomi's AI team tackles Automatic Speech Recognition (ASR) query errors by analyzing error patterns, employing BERT, ELECTRA and a soft‑masked BERT model, generating synthetic noisy data with a fuzzy‑phoneme generator, and presenting experimental results and future research directions.
This article presents a comprehensive overview of how 58.com leverages large‑scale voice data from call‑center, private phone, and micro‑chat platforms, detailing data collection, annotation, Kaldi‑based chain model training, lattice‑free techniques, and end‑to‑end Transformer‑CTC models to improve Chinese speech recognition performance.
The article presents a comprehensive technical overview of 58.com’s intelligent voice analysis platform, detailing its business background, system architecture, speech and NLP technologies, speaker diarization methods, model performance, data labeling workflow, and practical applications in call‑center quality inspection and user profiling.
The article details the design and deployment of an AI-powered intelligent voice quality inspection system at 58.com, covering its overall architecture, speech recognition, role identification, tag detection, rechecking platform, and backend infrastructure, and demonstrates its impact on call‑center efficiency and service quality.
This article presents a comprehensive overview of automatic speech recognition (ASR) error correction techniques employed by Xiaomi's Xiao‑Ai team, detailing problem definition, related work on BERT and ELECTRA, a custom generator‑discriminator architecture with a fuzzy‑phoneme simulator, experimental results, and prospective research directions.
This article presents the engineering architecture of intelligent voice robots, detailing voice preprocessing, intent recognition, slot extraction, dialogue management, and showcases multiple enterprise use cases that improve efficiency and revenue across sales, customer service, and recruitment.
This article provides a comprehensive technical overview of modern speech recognition, covering Didi’s driver‑assistant and smart‑customer‑service applications, fundamental ASR concepts, classic GMM‑HMM methods, deep‑learning breakthroughs such as DNN‑HMM, CTC, attention‑based and transformer models, practical training tricks, signal‑processing steps, and multimodal fusion techniques.
In this interview, Didi AI Labs’ Han Kun explains how the DELTA platform unifies TensorFlow‑based NLP and speech models—supporting tasks from text classification to voice emotion recognition—through a modular, easily deployable architecture, accelerating development, powering Didi products, and now open‑sourced for broader AI collaboration.
This article provides a comprehensive overview of AI's rapid evolution, covering deep learning foundations, machine learning components, natural language processing advances, speech recognition breakthroughs, multimodal interaction, computer vision progress, model compression techniques, and the shift from data‑driven to knowledge‑based AI approaches.
This article presents a comprehensive overview of 360's voice interaction platform, detailing dialogue system fundamentals, platform architecture, and core technologies such as semantic understanding, dialog management, and question answering, all driven by deep learning and multimodal innovations.
The article reviews Interspeech 2019, summarizing major trends and representative papers in end‑to‑end speech recognition, synthesis, natural language understanding, speaker recognition, and speech translation, while also highlighting best student papers and providing resources for further study.
This article describes the TPNN deep‑learning platform’s multi‑GPU acceleration, data‑parallel BMUF training, LSTM‑CTC acoustic modeling, and a suite of mobile‑side optimizations—including model pruning, 8‑bit quantization, low‑precision matrix multiplication and mixed‑precision computation—that together achieve over 92% recognition accuracy for children’s English speech on both server and mobile devices.
This article presents a comprehensive overview of modern speech recognition technology, covering basic ASR concepts, classic acoustic and language models, deep‑learning approaches such as DNN‑HMM, CTC, attention‑based and transformer models, multimodal fusion, signal‑processing pipelines, and practical deployment considerations at Didi.
Microsoft appointed its first Chinese Global Technical Fellow, Huang Xuedong, as the company’s Global AI CTO, overseeing Azure’s speech, translation, vision, and language services, while highlighting his groundbreaking achievements such as achieving human‑level word error rates and leading AI research teams.
The talk introduced iQIYI’s HomeAI platform, which combines user profiling (including voiceprint and age detection) with automatic video semantic extraction to enable natural, multi‑turn voice‑based video search—addressing hot‑content updates, contextual awareness, device environments, and personalized recommendations for screen‑less or accessibility‑focused users.
This article presents a complete solution for 58.com’s human‑machine voice dialogue robot, detailing its background, overall architecture, intelligent outbound process, core functions such as call service, anti‑spam, status recognition, multi‑turn dialogue management, intent classification, slot extraction, whole‑round intent detection, and various practical application scenarios.
This article presents the end‑to‑end solution of 58.com’s voice dialogue robot, detailing its overall architecture, intelligent outbound process, core functions such as call dialing, status recognition, dialogue management, intent detection, and showcasing multiple real‑world application scenarios that improve sales, operations, and customer service efficiency.
This article presents a distributed system that efficiently supports large‑scale n‑gram language models for automatic speech recognition by introducing caching, a two‑level distributed index, batch processing, and a cascading fault‑tolerance mechanism, demonstrating robust scalability and low communication overhead in Tencent's WeChat ASR service.
Intelligent speech recognition converts spoken audio to text using a pipeline of feature extraction, acoustic and language modeling, where deep neural networks—especially CNN, LSTM, and hybrid CLDNN architectures—drive high accuracy, enabling mobile voice input, call‑center transcription, legal record keeping, and Tencent Cloud ASR’s 97% Mandarin accuracy with speaker separation and on‑premises deployment.
This article presents a comprehensive overview of an intelligent voice robot's dialogue management system, detailing its architecture, natural language understanding components, dialogue manager design, strategy handling, and workflow processes to achieve fluent multi‑turn interactions in telephone scenarios.
At ACL 2019, Didi unveiled DELTA, an open‑source TensorFlow‑based training framework that unifies NLP and speech tasks, offers configurable pipelines, benchmark models, and seamless deployment, enabling AI developers to quickly move from research to production while leveraging Didi’s extensive open‑source ecosystem.
This article introduces Alibaba's new e‑book collection of five ICASSP‑accepted papers that showcase advances in speech recognition, synthesis, and emotion detection, detailing novel models like DFSMN, A‑LSTM, and speaker‑adaptation techniques that dramatically improve speed, size, and accuracy.
Using AI to automatically understand and secure audio content, this article discusses the challenges of manual audio analysis, outlines a four‑step pipeline—audio segmentation, speech‑to‑text, labeling, and synthesis—and describes models such as VAD, ASR, sound classification, text recognition, and behavior detection for live‑stream moderation.
This article reviews the fundamentals of automatic speech recognition, compares traditional MFCC‑GMM‑HMM pipelines with modern deep neural network approaches such as DNN‑HMM, LSTM‑CTC, and attention‑based models, and illustrates each evolution step with flowchart diagrams and key references.
