52 results about "Speech Acoustics" patented technology

To achieve an improvement in recognition performance, a non-speech acoustic model correction unit adapts a non-speech acoustic model representing a non-speech state using input data observed during an interval immediately before a speech recognition interval during which speech recognition is performed, by means of one of the most likelihood method, the complex statistic method, and the minimum distance-maximum separation theorem.

A method is developed which includes 1) defining a switching state space model for a continuous valued hidden production-related parameter and the observed speech acoustics, and 2) approximating a posterior probability that provides the likelihood of a sequence of the hidden production-related parameters and a sequence of speech units based on a sequence of observed input values. In approximating the posterior probability, the boundaries of the speech units are not fixed but are optimally determined. Under one embodiment, a mixture of Gaussian approximation is used. In another embodiment, an HMM posterior approximation is used.

Acoustic signals are analyzed by two-dimensional (2-D) processing of the one-dimensional (1-D) speech signal in the time-frequency plane. The short-space 2-D Fourier transform of a frequency-related representation (e.g., spectrogram) of the signal is obtained. The 2-D transformation maps harmonically-related signal components to a concentrated entity in the new 2-D plane (compressed frequency-related representation). The series of operations to produce the compressed frequency-related representation is referred to as the “grating compression transform” (GCT), consistent with sine-wave grating patterns in the frequency-related representation reduced to smeared impulses. The GCT provides for speech pitch estimation. The operations may, for example, determine pitch estimates of voiced speech or provide noise filtering or speaker separation in a multiple speaker acoustic signal.

A client-server architecture for Automatic Speech Recognition (ASR) applications, includes: (a) a client-side including: a client being part of distributed front end for converting acoustic waves to feature vectors; VAD for separating between speech and non-speech acoustic signals; adaptor for WebSockets; and (b) a server side including: a web layer utilizing HTTP protocols and including a Web Server having a Servlet Container; an intermediate layer for transport based on Message-Oriented Middleware being a message broker; a recognition server and an adaptation server both connected to said intermediate layer; a Speech processing server; a Recognition Server for instantiation of a recognition channel per client; an Adaptation Server for adaptation acoustic and linguistic models for each speaker; a Bidirectional communication channel between a Speech processing server and client side; and a Persistent layer for storing a Language Knowledge Base connected to said Speech processing server.

The invention provides a formant-frequency-based Mandarin single final voice visualizing method. The method comprises the following steps: framing and windowing an original voice signal; extracting short-time energy, formant frequency and fundamental tone frequency of each frame of signal; correcting errors of specific values of the formant frequency and the fundamental tone frequency by adoptinga median smoothing method; mapping different pronunciations into different color aspects by utilizing the formant frequency, and correcting; reflecting variation tendency of pronunciation time, energy and fundamental tone frequency in an image; and differentiating different Mandarin single final pronunciations by colors. The method is easy to implement by only extracting acoustic phonetic parameters of short-time energy, formant frequency and fundamental tone frequency of a voice signal; soft decision is introduced, each pronunciation is not subject to hard decision, but represented by different colors, and visualizing effect of different speakers on the same pronunciation is based on the principle of seeking common grounds while reserving differences, so that the decision on pronunciation more accords with subjective perception of people.

The invention discloses a multiple self-adaption based model compensation type speech recognition method. The method includes firstly, subjecting parameters of a clean speech acoustic model trained in a training environment in advance to transformation so as to obtain a noisy speech acoustic model matched with a practical testing environment; then, taking the noisy speech acoustic model obtained through first self-adaption as a base environment acoustic model, constructing a transformation relation between noisy speech corresponding to the base environment acoustic model and noisy testing speech of a practical environment, and subjecting the base environment acoustic model to model self-adaption again, wherein the model self-adaption includes positive self-adaption and negative self-adaption; finally, comparing output likelihood values of the positive self-adaption and the negative self-adaption, and taking the noisy speech acoustic model with the larger likelihood value as a model self-adaption result. According to the multiple self-adaption based model compensation type speech recognition method, model self-adaption precision can be further improved so as to obtain the noisy speech acoustic model better matched with the practical testing environment.

The invention discloses a method and system for speech recognition of dysarthria based on visual facial contour movement. The system includes multi-modal data acquisition, multi-modal fusion feature calculation, multi-modal speech recognition calculation and language model calculation modules; multi-modal The state data acquisition calculation module is used to obtain the facial contour motion video data of the dysarthria and the voice data synchronized with the video; the multi-modal fusion feature calculation module is used to fuse the facial contour motion features and speech acoustic features; multi-modal speech recognition The calculation module is used to obtain the mapping relationship from multimodal features to phoneme characters; the language model calculation module is used to obtain the mapping relationship from phoneme characters to Chinese sentences. The present invention obtains the fused multimodal features by fusing the speech acoustic feature parameters and the pronunciation actions of the dysarthria, and utilizes the fused multimodal features to perform dysarthria speech recognition, thereby effectively improving dysarthria speech recognition Accuracy.

The present application discloses a method, device, computer equipment and storage medium for establishing a voiceprint model, wherein the method includes: dividing the input target user's voice signal into frames, and extracting the voice acoustic features of the framed voice signals respectively; A plurality of said speech acoustic features are input into a deep learning model based on neural network training, and are assembled into at least one cluster structure; calculating the mean value and standard deviation of at least one said cluster structure; performing coordinate transformation on said mean value and standard deviation and calculating the activation function to obtain feature vector parameters; inputting the feature vector parameters and the identity verification result of the target user into a preset basic model to obtain a voiceprint model corresponding to the target user. The speech acoustic features extracted in this application are based on the cluster structure obtained in the deep neural network training, and then the cluster structure is subjected to coordinate mapping and activation function calculation to obtain a voiceprint model, which can reduce the voice recognition error rate of the voiceprint model.

The present invention relates to the technical fields of speech synthesis, speech recognition, and sound cloning. The present invention combines speech synthesis technology, speech recognition technology, and transfer learning technology to provide a sound cloning implementation scheme based on Bottleneck features (language features of audio), including a training system and training methods; use a small number of samples to provide TTS services with high naturalness and similarity, so as to provide TTS services with target user characteristics, and solve the problems of large sample size, long production cycle, and high labor cost of speech synthesis technology services. The training system includes: a data acquisition module, an acoustic feature extraction module, a speech recognition module, a prosody module, a multi-person speech acoustic module, and a speech synthesis module; the present invention also provides a training method based on the above-mentioned system, including preparing training corpus, acoustic feature extraction , training and fine-tuning of each module, and speech synthesis.