Recently, a lot of deep learning model successful in taking over conventional methods in speech processing fields. Vector quantization is a popular technique to reduce the amount of speech data before transmitting. The conventional vector quantization method is based on the mathematical model. Last few years, the Vector Quantized Variational AutoEncoder has been proposed for an end-to-end vector quantization based on deep learning techniques. In this paper, we investigate the sub-band quantization in the Vector Quantized Variational AutoEncoder. This model can concentrate on specific frequency bands to assign more bits and leave the unnecessary band with few bits. Experimental results show the efficiency of the proposed quantization method for the spectral envelope parameters of the high-quality vocoder that operates at 48 kHz sampling frequency named WORLD vocoder. At the same four target bit rates, the sub-band Vector Quantized Variational AutoEncoder can reduce the Log Spectral Distortion around 0.93 dB in average.

This paper presents a new technique using animated texts as the speech features' visualization medium for checking and detecting language learners' pronunciation. The proposed visualization tool will transform learners' speech features such as pitch, tempo or rhythm into animated texts form, and the mispronounce parts can be located by comparing them with the correct sample. In our previous experiments, Japanese language learners gave positive feedback on the animated texts designs as their speech visualization tool. By practicing this tool, learners found that the mispronounce parts in their speech can be detected and confirmed easily. However, in order to have more practical feedback and response from the learners, besides using the plain speech data set, which only express random speech contents, the daily conversation speech data set is proposed as the data sample. In this paper, the comparison between both database samples for determining the proposed visualization tool's approachability was observed. Evaluation experiments results showed that participants gave positive responses to the animated texts visualization tool and were able to understand speech features in the visualized texts form better by using the daily conversation as the speech sample.

This paper proposes a speech segment selection method based on machine learning for concatenative speech synthesis systems. The proposed method has two novel features. One is its use of support vector machine (SVM) to estimate the subjective correctness of pitch accent with respect to each accent phrase of possible candidate speech segments. The other is its use of a determination function to identify the best segment based on SVM output. The determination function involves two assessments; one counts the number of each sign of SVM output and the other compares the distance values. The sign of SVM output is generally used to classify target objects, but the distance SVM output also represents important information. An experiment that assesses SVM performance for Japanese accent validity shows that its accuracy is 81%. To confirm the effectiveness of the proposed segment selection method, preference tests are conducted. The test indicates that the proposed method can yield Japanese synthesized speech with more natural intonation than the conventional method that uses only target cost and concatenation cost.

This paper describes a new recording script generation method that can create speech databases for corpus-based TTS (text to speech) systems. This method is efficient due to its two features: (1) it has a 2-stage algorithm to generate the recording script with consideration of the balance of triphone, syllable and morpheme elements; (2) it can control types of phonetic elements included in the recording script via the weight coefficients of the phonetic elements. An evaluation shows that the 2-stage algorithm is effective in raising the coverage of phonetic elements and that this method yields a recording script containing various phonetic elements. A preference test shows that changing the selection criteria influences the quality of the synthesized speech. The same test also shows that it is better to take account of morpheme-based elements than syllable-based elements in generating a task-specific recording script.

A speech-coding method called “recovery sub-codec (RSC)” is described that improves speech quality under packet-loss conditions, which can occur even in high-speed wide-band networks. The RSC uses a sub bit-stream as additional information in the next speech frame in addition to the main bit-stream in the current frame. The main bit-stream is coded using a low-comp ression high -quality main coder and the sub bit-stream is coded using a high-compression sub-coder. The internal states of the recovery sub-coder can be updated using the output from the high-quality main coder, so they are independent of previous frames and are not affected by previous packet loss. Subjective quality tests of several speech codecs based on the proposed method show that it significantly improves the concealment of packet loss, both for narrow- and wide-band speech and audio signals.

This paper presents the 4 kbit/s speech coding candidate submitted by AT&T, Conexant, Deutsche Telekom, France Telecom, Matsushita, and NTT for the ITU-T 4 kbit/s selection phase. The algorithm was developed jointly based on the qualification version of Conexant. This paper focuses on the development carried out during the collaboration in order to narrow the gap to the requirements in an attempt to provide toll quality at 4 kbit/s. This objective is currently being verified in independent subjective tests coordinated by ITU-T and carried out in multiple languages. Subjective tests carried out during the development indicate that the collaboration work has been successful in improving the quality, and that meeting a majority of the requirements in the extensive selection phase test is a realistic goal.

Speech coding at very low bit-rate is useful for purposes such as voice communication over computer networks. However speech coding at around 2.0 kbit/s is difficult for CELP coders while maintaining a high quality. In this paper, a speech coding model called 'normalized pitch waveform' and its quantization scheme are presented, aiming for effective compression coding of the 'voiced' speech. Listening tests have proven that an efficient and high quality coding has been achieved at 2.0 kbit/s, less than half of the FS1016. Furthermore this paper discusses the disadvantage of the normalized pitch waveform and presents an alternative method of using non-normalized pitch waveform. Encoding of a transitional 'mixed' state between the 'voiced' and the 'unvoiced' state is discussed for further improvements.

This paper describes the design of a toll-quality 4-kbit/s speech coder based on phase-adaptive PSI-CELP. This adaptation method not only gives pitch periodicity to the random excitation but also synchronizes the basic point of the stored random vector with the pitch phase. We further improve the proposed coder by introducing a backward gain prediction scheme. In subjective evaluation experiments, there is no significant difference between the quality of ITU-T G.726 32-kbit/s coder and that of the proposed 4-kbit/s coder under the conditions of normal and low input levels, tandem connection for clean speech. In noisy environments, there are also no significant differences between G.726 and 4-kbit/s coders from MOS results of the ACR test.

Low-cost highly-efficient speech coding is important for personal multimedia communications. In this paper we propose a new low-complexity speech coding method called "Dual-Pulse CS-CELP (DP-CS-CELP)" at 7.8 kbit/s. This method is based on ITU-T G.729. To reduce the complexity, we applied a new excitation model to the random code vectors and simplified the LPC coding, adaptive codebook search, perceptual weighting, and the other structures. The number of operations for this encoder is 3.80 MOPS, which achieves real-time speech coding and decoding on personal computers. Although the efficiency of this coder is a little lower than that of G.729, the MOS listening test showed that the subjective quality was equivalent to or a little better than that of G.726 (32-kbit/s ADPCM).

This paper describes the design of a speech coder called pitch synchronous innovation CELP (PSI-CELP) for low hit-rate mobile communications. PSI-CELP is based on CELP, but has more adaptive excitation structures. In voiced frames, instead of conventional random excitation vectors, PSI-CELP converts even the random excitation vectors to have pitch periodicity by repeating stored random vectors as well as by using an adaptive codebook, in silent, unvoiced, and transient frames, the coder stops using the adaptive codebook and switches to fixed random codebooks. The PSI-CELP coder also implements novel structures and techniques: an FIR-type perceptual weighting filter using unquantized LPC parameters, a random codebook with a conjugate structure trained to be robust against channel errors, codebook search with delayed decision, a gain quantization with sloped amplitude, and a moving average prediction coding of LSP parameters, Our speech coder is implemented by DSP chips. Its coded speech quality at 3.6 kb/s with 2.0 kb/s redundancy is comparable to that of the Japanese full-rate VSELP coder at 6.7 kb/s with 4.5 kb/s redundancy. The basic structure of this PSI-CELP coder has been chosen as the Japanese half-rate speech codec for digital cellular telecommunications.>

This paper proposes high-quality and low bit-rate (3.6 and 2.4 kbit/s) coders using a pitch synchronous innovation CELP (PSI-CELP) method or a phase adaptive PSI-CELP. PSI-CELP, which is used as the excitation structure of the half-rate codec for the standard of Japanese digital mobile telephony, is based on CELP but adds pitch synchronous innovation, which means that even random codevectors are adaptively converted to have pitch periodicity for voiced frames. Phase adaptive PSI-CELP makes not only the periodicity, like in PSI-CELP, but also the phase of random codevectors equal to those of an adaptive codevector. The subjective qualities of the 3.6- and 2.4-kbit/s coders exceed those of the 6.7-kbit/S VSELP coder, which is the full-rate codec for the standard of Japanese digital mobile telephony, and-the 4.8-kbit/s U.S. Federal Standard 1016 CELP coder, respectively, in the error-free condition.>

A 4.8-kb/s delayed decision code excited linear prediction (CELP) coder that uses tree coding is described. In conventional CELP coding, short-term and long-term prediction parameters as well as excitation parameters are sequentially determined. In the proposed delayed decision CELP coding, a tree coding method is utilized. The long-term prediction and excitation parameter candidates obtained in each subframe are listed as a tree and the optimum combined parameter sequences are selected to minimize global quantization distortion over the coding frame. The proposed coding method significantly increases the quality of the 4.8-kb/s CELP coder at the cost of an additional 5-ms coding delay.>

This paper describes an approach for a phoneme recognition based on a clustering method which considers phonemic features in each frame. In the clustering, both acoustic and phonemic features of speech are used. The acoustic features are LPC cepstral coefficients, the cepstral changes between adjacent frames and the power changes. The combination of these features shows both the static and dynamic properties of the spectrum. The phonemic feature in a frame is composed of a triplet of phonemic symbols. A vector quantization method is applied for the clustering. Experiment of extracting phonemic label sequences is performed, considering a contiguity of code sequences between the input and the reference phonemic patterns.