0 views collected around this technical thread.
ICBC's remote‑banking hotline system uses AI, speech recognition and Python keyword extraction to rank inbound business volumes and surface hot‑word trends, delivering early alerts that help prevent risks, resolve customer issues, and support data‑driven decision making across millions of daily transactions.
This article explains how to use NVIDIA Riva to rapidly deploy speech AI services, covering Riva's overview, Chinese ASR model updates, TTS capabilities, customization options, the Quickstart tool, and a Q&A session that clarifies deployment, model fine‑tuning, and integration with NeMo and Triton.
This article introduces NVIDIA’s open‑source NeMo framework, detailing its PyTorch‑based architecture for Speech AI, ASR and TTS training, NLP and LLM support, GPU‑optimized parallelism, pre‑trained model resources, fine‑tuning techniques, and the accompanying NeMo Aligner and Framework tools.
This article explains how to efficiently deploy speech AI models—including ASR and TTS—on GPUs using NVIDIA's Triton Inference Server and TensorRT, covering background challenges, GPU‑based solutions, decoding optimizations, Whisper acceleration with TensorRT‑LLM, streaming TTS improvements, voice‑cloning pipelines, future plans, and a Q&A session.
This article details the design, evolution, and best‑practice implementation of load‑balancing for ASR (speech‑recognition) services in Ctrip's massive call‑center, covering component architecture, MRCP integration, challenges with traditional balancers, and two practical solutions using FreeSWITCH distributor and OpenSIPS.
The article discusses cloud-based audio editing tool architecture, focusing on text‑based editing enabled by ASR, hierarchical DOM (Word, Sentence, Paragraph), performance challenges with massive character nodes, and optimizations like viewport‑based rendering and efficient drag‑select, achieving large speed gains for long recordings.
The authors present a multimodal system that automatically extracts high‑energy video segments for dynamic covers by analyzing subtitles, audio, visual frames, and danmu, employing LLM prompt‑tuning, scene‑cut detection, and aesthetic scoring to reduce manual effort and boost click‑through rates.
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 article reviews the technical background of simultaneous speech translation, compares offline and real‑time scenarios, details ASR and MT technologies, describes the system architecture and design strategies, and discusses the major challenges and solutions for deploying robust, low‑latency translation services.
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 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.
This article provides a comprehensive introduction to NVIDIA's NeMo toolkit for conversational AI, detailing its ASR capabilities, model architectures, training workflow, Chinese language support, deployment options, and additional speech AI features such as VAD and speaker diarization.
This article presents comprehensive best‑practice guidelines for deploying conversational speech AI—including ASR and TTS pipelines—on GPU servers using NVIDIA Triton Inference Server and TensorRT, covering workflow overview, performance optimizations, streaming inference, and real‑world deployment tips.
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 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.
This article reviews the IWSLT 2022 competition tasks, explains Huawei’s cascade offline speech‑to‑text translation pipeline, details four major technical innovations—including ensemble‑based ASR de‑noise, context‑aware re‑ranking, domain‑controlled training, and length‑control strategies—and presents experimental results that demonstrate Huawei’s leading performance across multiple language directions.
This article reviews Huawei Translation’s top-ranking results in the IWSLT 2022 speech translation competition across speech‑to‑speech, offline speech‑to‑text, and length‑controlled translation tasks, and details their cascade and end‑to‑end technical approaches, including domain‑controlled ASR, context‑aware MT re‑ranking, and VITS‑based TTS.
This article presents an in‑depth overview of Alibaba's real‑time voice dialogue system, covering the Hotline XiaoMi robot, the unique challenges of spoken interactions such as colloquialism, multimodality and duplex communication, and the research advances in ASR‑robust SLU, emotion detection, colloquial processing, and duplex conversation modeling.
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.
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.