JPH08278800A - Voice communication system - Google Patents
Voice communication systemInfo
- Publication number
- JPH08278800A JPH08278800A JP7080034A JP8003495A JPH08278800A JP H08278800 A JPH08278800 A JP H08278800A JP 7080034 A JP7080034 A JP 7080034A JP 8003495 A JP8003495 A JP 8003495A JP H08278800 A JPH08278800 A JP H08278800A
- Authority
- JP
- Japan
- Prior art keywords
- band
- signal
- prediction
- wide
- prediction coefficient
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Landscapes
- Transmission Systems Not Characterized By The Medium Used For Transmission (AREA)
Abstract
Description
【産業上の利用分野】本発明は音声通信システムに関
し、特に電話回線等の伝送路を狭帯域音声信号で伝送す
る音声通信システムに関するものである。BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice communication system, and more particularly to a voice communication system for transmitting a narrow band voice signal through a transmission line such as a telephone line.
【０００１】ビデオフォン、電話会議システム、テレビ
会議システム等のマルチメディア通信システムにおける
音声通信では、電話回線を用いて電話帯域（３００〜３
４００Hz）の音声を伝送するか、ＩＳＤＮ回線のような
広帯域の伝送路を使用する場合は、広帯域（５０〜７０
００Hz）な音声を符号化（例えばＩＴＵ−ＴＧ．７２２
の６４kb/s符号化）してディジタル伝送を行っている。In voice communication in a multimedia communication system such as a videophone, a telephone conference system, a video conference system, etc., a telephone band (300 to 3) is used by using a telephone line.
When transmitting voice of 400Hz or using a wideband transmission line such as an ISDN line, a wideband (50 to 70)
00 Hz) voice is encoded (for example, ITU-TG.722).
64 kb / s encoding) for digital transmission.
【０００２】この場合、より廉価で多くの回線を設定す
るためには、前者の電話回線を用いた狭帯域音声通信が
必要となる。In this case, narrow band voice communication using the former telephone line is required in order to set up many lines at a lower cost.
【０００３】[0003]
【従来の技術】図４には電話回線を用いた従来から周知
の狭帯域音声通信システムの構成が示されており、ハン
ドセットＴＭ（送話器）からの音声信号は狭帯域音声送
信器１で狭帯域化されて電話回線の伝送路２へ送出され
る。2. Description of the Related Art FIG. 4 shows the configuration of a conventionally well-known narrow band voice communication system using a telephone line, and a voice signal from a handset TM (speaker) is transmitted by a narrow band voice transmitter 1. The band is narrowed and transmitted to the transmission line 2 of the telephone line.
【０００４】伝送路２からの狭帯域音声信号は狭帯域音
声受信器３で受信され、ハンドセットＲＶ（受話器）か
ら出力される。The narrow band voice signal from the transmission line 2 is received by the narrow band voice receiver 3 and output from the handset RV (handset).
【０００５】このように、ハンドセットを用いて音声の
みの通信を行う電話では、狭帯域音声でも大きな不都合
を感じないが、近年では、ビデオフォン、電話会議シス
テム、テレビ会議システム等、画像を見ながら対話や会
議を行うマルチメディア通信システムが普及して来てお
り、このようなマルチメディア通信システムにおける音
声通信、特にスピーカを用いて受聴する場合では、電話
帯域の音声での通信は臨場感や自然性に乏しく感じるよ
うになる。このため、通常の電話で用いられている帯域
より広帯域の音声を用いることが望ましい。As described above, a telephone that performs voice-only communication using a handset does not feel a great inconvenience even with narrow-band voice, but in recent years, videophones, telephone conference systems, video conference systems, etc., can be used while viewing images. 2. Description of the Related Art Multimedia communication systems for conducting dialogues and conferences have become widespread, and voice communication in such multimedia communication systems, especially when listening using a speaker, communication using voice in the telephone band is realistic and natural. I feel less sexual. For this reason, it is desirable to use voice with a wider band than that used in a normal telephone.
【０００６】一方、ＩＳＤＮ回線のような広帯域の伝送
路を用いて広帯域音声を直接伝送する場合はこのような
問題が無いが、回線料金が通常の電話回線より高くな
り、また音声のディジタル化および圧縮のための音声符
号器・復号器を備える必要があるため、通信コストが高
くなる。On the other hand, when broadband voice is directly transmitted using a broadband transmission line such as an ISDN line, there is no such problem, but the line charge is higher than that of a normal telephone line, and the digitization of voice and Since it is necessary to provide a voice encoder / decoder for compression, the communication cost becomes high.
【０００７】そこで、図５に示すように受信側において
狭帯域音声受信器３から出力される狭帯域音声信号を広
帯域化処理部４で広帯域化してスピーカＳＰから出力さ
せる方式が提案されるに到っている。Therefore, as shown in FIG. 5, a method has been proposed in which a narrow band audio signal output from the narrow band audio receiver 3 on the receiving side is widened by the wide band processing unit 4 and output from the speaker SP. ing.
【０００８】この場合の広帯域化処理部４としては、線
形変換を用いた方式が提案されている。In this case, as the band widening processing section 4, a method using linear conversion has been proposed.
【０００９】[0009]
【発明が解決しようとする課題】しかしながら、このよ
うな従来の広帯域化処理部は、広帯域音声信号の復元精
度があまり高くないという問題点があった。However, such a conventional wide band processing unit has a problem that the restoration accuracy of the wide band voice signal is not so high.
【００１０】従って、本発明は、電話回線等の伝送路か
らの狭帯域音声信号を広帯域化処理部で広帯域化して出
力する音声通信システムにおいて、広帯域化処理音声の
復元精度を向上することを目的とする。Therefore, it is an object of the present invention to improve the restoration accuracy of wide band processed voice in a voice communication system in which a narrow band voice signal from a transmission line such as a telephone line is widened by a wide band processing unit and output. And
【００１１】[0011]
【課題を解決するための手段】上記の目的を達成するた
め、本発明に係る音声通信システムにおいては、広帯域
化処理部が、狭帯域受話音声信号をアナログ／デジタル
変換する変換器と、該変換器の出力信号に対して線形予
測分析分析を行うことにより狭帯域予測係数を求める線
形予測分析部と、該変換器の出力信号及び該狭帯域予測
係数から狭帯域予測残差信号を求める逆フィルタと、該
狭帯域予測係数から広帯域予測係数を推定するニューラ
ルネットワーク部と、該狭帯域予測残差信号に対して非
線形演算を施して広帯域予測誤差信号を発生させる非線
形処理部と、該広帯域予測係数を係数とし該広帯域予測
誤差信号を入力信号とする合成フィルタと、該変換器の
出力信号の第１の周波数帯域を通過させる第１の帯域通
過フィルタと、該合成フィルタの出力信号の第２及び第
３の周波数帯域をそれぞれ通過させる第２及び第３の帯
域通過フィルタと、該第１乃至第３の帯域通過フィルタ
の出力信号を入力して広帯域音声信号を合成する合成部
と、を備えている。In order to achieve the above object, in a voice communication system according to the present invention, a wide band processing section includes a converter for analog / digital converting a narrow band received voice signal, and the conversion. Prediction analysis unit for obtaining a narrow band prediction coefficient by performing a linear prediction analysis analysis on the output signal of the converter, and an inverse filter for obtaining a narrow band prediction residual signal from the output signal of the converter and the narrow band prediction coefficient A neural network unit that estimates a wideband prediction coefficient from the narrowband prediction coefficient, a nonlinear processing unit that performs a nonlinear operation on the narrowband prediction residual signal to generate a wideband prediction error signal, and the wideband prediction coefficient A coefficient and a wide band prediction error signal as an input signal, a first band pass filter for passing a first frequency band of the output signal of the converter, The second and third band pass filters for passing the second and third frequency bands of the output signal of the synthesis filter, respectively, and the output signals of the first to third band pass filters are input to generate a wide band audio signal. And a synthesizing unit for synthesizing.
【００１２】また、上記のニューラルネットワーク部
は、該狭帯域予測係数から低域部予測係数及び高域部予
測係数をそれぞれ推定する第１及び第２のニューラルネ
ットワーク部で構成することができ、該合成フィルタ
は、該低域部予測係数及び高域部予測係数をそれぞれ係
数とし該広帯域予測誤差信号を入力信号とし、各出力を
それぞれ第２及び第３の帯域通過フィルタに与える第１
及び第２の合成フィルタで構成することができる。The neural network section may be composed of first and second neural network sections for estimating a low band prediction coefficient and a high band prediction coefficient, respectively, from the narrow band prediction coefficient. The synthesis filter uses the low band prediction coefficient and the high band prediction coefficient as coefficients, receives the wide band prediction error signal as an input signal, and outputs the outputs to the second and third band pass filters, respectively.
And a second synthesis filter.
【００１３】また、上記の非線形処理部は、全波整流、
半波整流、又は二乗演算を用いることができる。Further, the above-mentioned non-linear processing unit is a full-wave rectifier,
Half-wave rectification, or squaring operations can be used.
【００１４】[0014]
【作用】本発明において、伝送路には狭帯域の音声信号
を伝送し、受信側において設けた広帯域化処理部が受信
音声信号の帯域拡張を行って再生を行う。In the present invention, a narrow band audio signal is transmitted to the transmission path, and the band widening processing unit provided on the receiving side expands the band of the received audio signal to reproduce it.
【００１５】この広帯域化処理部では、アナログ／デジ
タル変換器でアナログ信号からデジタル信号に変換され
た狭帯域受話音声に対して線形予測分析部で線形予測分
析を行い狭帯域予測係数を求め、逆フィルタにより狭帯
域予測残差信号を求める。In this wide band processing unit, the linear prediction analysis unit performs linear prediction analysis on the narrow band received voice converted from the analog signal to the digital signal by the analog / digital converter to obtain the narrow band prediction coefficient, and the inverse The narrow band prediction residual signal is obtained by the filter.
【００１６】狭帯域予測係数は、これを入力とするニュ
ーラルネットワーク部により広帯域の予測係数の推定を
行う。一方、狭帯域予測残差信号に対しては、これに非
線形処理部で絶対値演算（全波整流）、半波整流、又は
二乗演算等の非線形操作を行うことにより高調波成分を
発生させて広帯域の予測残差信号を生成する。The narrow band prediction coefficient is used as an input to estimate the wide band prediction coefficient by the neural network unit. On the other hand, for the narrowband prediction residual signal, the nonlinear processing unit performs nonlinear operations such as absolute value calculation (full-wave rectification), half-wave rectification, or square calculation to generate harmonic components. Generate a wideband prediction residual signal.
【００１７】この広帯域予測残差信号をニューラルネッ
トワーク部からの広帯域予測係数を用いて合成フィルタ
で再び線形予測合成し、その低域周波数成分および高域
周波数成分をそれぞれ帯域通過フィルタから取り出し、
元の狭帯域音声デジタル信号から帯域通過フィルタによ
り取り出された中域周波数成分に合成部で加えることに
より広帯域音声信号を生成する。This wideband prediction residual signal is again linearly predicted and synthesized by a synthesis filter using the wideband prediction coefficient from the neural network unit, and its low frequency component and high frequency component are respectively taken out from the band pass filter,
A wideband speech signal is generated by adding the middle frequency component extracted from the original narrowband speech digital signal by the bandpass filter in the synthesis unit.
【００１８】このように本発明では、狭帯域音声の予測
残差信号に対して非線形処理を施し、このときに発生す
る高調波成分を利用することにより、少ない演算量で帯
域を増加させることなく広帯域音声信号の再生を行うこ
とができ、受話音声品質の改善が図れる。As described above, according to the present invention, the non-linear processing is performed on the prediction residual signal of the narrow band speech, and the harmonic components generated at this time are used to increase the bandwidth with a small amount of calculation. A wideband voice signal can be reproduced, and the quality of received voice can be improved.
【００１９】[0019]
【実施例】図１は本発明に係る広帯域音声通信システム
における広帯域処理部の実施例を示しており、この実施
例では、狭帯域受話音声信号をアナログ／デジタル変換
する変換器（Ａ／Ｄ変換器）４１と、このＡ／Ｄ変換器
４１からのデジタル狭帯域音声信号に対して線形予測分
析分析を行うことにより狭帯域予測係数を求める線形予
測分析部４２と、Ａ／Ｄ変換器４１からのデジタル狭帯
域音声信号及び線形予測分析部４２から得られる狭帯域
予測係数から狭帯域予測残差信号を求める逆フィルタ４
３と、線形予測分析部４２から得られる狭帯域予測係数
より広帯域予測係数を推定するニューラルネットワーク
部４４と、逆フィルタ４３で得られた狭帯域予測残差信
号に対して非線形演算を施して広帯域予測誤差信号を発
生させる非線形処理部４５と、ニューラルネットワーク
部４４で得られた広帯域予測係数を係数とし非線形処理
部４５で得られた広帯域予測誤差信号を入力とする合成
フィルタ４６と、Ａ／Ｄ変換器４１からのデジタル狭帯
域音声信号の第１の周波数帯域（３００〜３４００Hz）
を通過させる第１の帯域通過フィルタ４７と、該合成フ
ィルタの出力信号の第２及び第３の周波数帯域（５０〜
３００Hz，３４００〜７０００Hz）をそれぞれ通過させ
る第２及び第３の帯域通過フィルタ４８及び４９と、該
第１乃至第３の帯域通過フィルタ４７〜４９の出力信号
を合成して広帯域音声信号にする合成部５０と、で構成
されている。FIG. 1 shows an embodiment of a wide band processing unit in a wide band voice communication system according to the present invention. In this embodiment, a converter (A / D conversion) for analog / digital converting a narrow band received voice signal. 41), a linear prediction analysis unit 42 for obtaining a narrow band prediction coefficient by performing a linear prediction analysis analysis on the digital narrow band speech signal from the A / D converter 41, and the A / D converter 41. Inverse filter 4 for obtaining a narrow band prediction residual signal from the digital narrow band speech signal and the narrow band prediction coefficient obtained from the linear prediction analysis unit 42.
3, a neural network unit 44 that estimates a wide band prediction coefficient from the narrow band prediction coefficient obtained from the linear prediction analysis unit 42, and a wide band by performing a non-linear operation on the narrow band prediction residual signal obtained by the inverse filter 43. A non-linear processing unit 45 that generates a prediction error signal, a synthesis filter 46 that receives the wide band prediction error signal obtained by the non-linear processing unit 45 using the wide band prediction coefficient obtained by the neural network unit 44 as a coefficient, and an A / D First frequency band (300 to 3400 Hz) of the digital narrow band audio signal from the converter 41
A first band-pass filter 47 for passing the signal, and second and third frequency bands (50 to 50) of the output signal of the synthesis filter.
300 Hz, 3400 to 7000 Hz) and second and third band pass filters 48 and 49, respectively, and output signals of the first to third band pass filters 47 to 49 are combined into a wide band audio signal. And a section 50.
【００２０】この実施例の動作においては、受信側にお
いて再生した狭帯域音声信号を入力としてこれをＡ／Ｄ
変換器４１でＡ／Ｄ変換し、線形予測分析部４２では、
狭帯域デジタル信号に対して短時間区間毎に線形予測分
析を行い、狭帯域予測係数を求める。In the operation of this embodiment, the narrow band audio signal reproduced on the receiving side is input and is inputted to the A / D.
The converter 41 performs A / D conversion, and the linear prediction analysis unit 42
A narrow band prediction coefficient is obtained by performing a linear prediction analysis on the narrow band digital signal for each short time period.
【００２１】次にこの狭帯域予測係数を入力とするニュ
ーラルネットワーク部４４では、広帯域の予測係数を推
定する。このニューラルネットワーク部としては、例え
ば階層型ネットワークにより実現することができる。Next, the neural network unit 44, which receives the narrow band prediction coefficient as an input, estimates the wide band prediction coefficient. This neural network unit can be realized by, for example, a hierarchical network.
【００２２】図２には入力層と隠れ層と出力層から成る
３層ニューラルネットワーク部の構成例が示されてお
り、ネットワークの重み係数の学習には誤差逆伝搬法
（バックプロパゲーション法）等のアルゴリズムを用い
ることができる。FIG. 2 shows an example of the configuration of a three-layer neural network unit consisting of an input layer, a hidden layer, and an output layer. The error back propagation method (back propagation method) or the like is used for learning the weighting coefficient of the network. Can be used.
【００２３】このネットワークには線形予測分析部４２
からの狭帯域音声信号のＬＰＣケプストラム係数ｘ1 〜
ｘN を入力し、出力には広帯域のＬＰＣケプストラム係
数ｙ 1 〜ｙN が出力されるように重み係数の学習を行
う。また、推定に用いるパラメータとしてはＬＰＣケプ
ストラム係数以外にも反射係数等さまざまなものを用い
ることができる。In this network, the linear prediction analysis unit 42
LPC cepstrum coefficient x of the narrowband speech signal from1~
xNInput, and output is wideband LPC cepstrum
Number y 1~ YNLearning the weighting factors so that
U The parameters used for estimation are LPC caps.
In addition to the strum coefficient, various other factors such as the reflection coefficient are used.
Can be
【００２４】このスペクトルの変換関数は一般的には非
線形と考えられるため、ニューラルネットワーク部の適
用により線形変換を用いる場合より変換精度の向上が期
待できる。また、未学習入力に対する外挿効果も有す
る。Since the conversion function of this spectrum is generally considered to be non-linear, the conversion accuracy can be expected to be improved by applying the neural network unit as compared with the case where linear conversion is used. It also has an extrapolation effect on unlearned inputs.
【００２５】逆フィルタ４３は狭帯域入力音声信号に対
して線形予測分析部４２からの狭帯域予測係数を用いて
逆フィルタ処理を行い、狭帯域予測残差信号を求める。The inverse filter 43 performs an inverse filter process on the narrow band input speech signal using the narrow band prediction coefficient from the linear prediction analysis unit 42 to obtain a narrow band prediction residual signal.
【００２６】次にこの狭帯域予測残差信号に対してサン
プル毎に非線形処理部４５が非線形処理を施すことによ
り広帯域予測残差信号を生成する。これは絶対値演算の
ような非線形処理により高調波成分が発生することを利
用している。Next, the non-linear processing unit 45 performs non-linear processing on the narrow band prediction residual signal for each sample to generate a wide band prediction residual signal. This utilizes the fact that harmonic components are generated by non-linear processing such as absolute value calculation.
【００２７】また非線形処理部４５により広帯域化した
予測残差信号をニューラルネットワーク部４４からの広
帯域予測係数を係数とする予測合成フィルタ４６に通し
て広帯域音声信号を生成する。Further, the prediction residual signal whose band is widened by the non-linear processing section 45 is passed through a prediction synthesis filter 46 having a wide band prediction coefficient from the neural network section 44 as a coefficient to generate a wide band speech signal.
【００２８】この広帯域音声信号は、帯域通過フィルタ
４８及び４９を通すことにより、音声信号の低域周波数
成分（５０−３００Hz）及び高域周波数成分（３４００
−７０００Hz）をそれぞれ抽出する。This wide band audio signal is passed through band pass filters 48 and 49 to obtain a low frequency component (50-300 Hz) and a high frequency component (3400) of the audio signal.
-7000 Hz) is extracted.
【００２９】そして、中域周波数成分（３００−３４０
０Hz）が狭帯域入力信号から帯域通過フィルタ４７によ
り取り出されて合成部５０により帯域通過フィルタ４８
及び４９の低域周波数成分及び高域周波数成分に加え合
わせることで、広帯域音声信号（５０−７０００Hz）を
生成している。Then, the middle frequency components (300-340
0 Hz) is extracted from the narrow band input signal by the band pass filter 47 and is combined by the band synthesizer 50.
And 49, the low-frequency component and the high-frequency component are added together to generate a wideband audio signal (50-7000 Hz).
【００３０】図３は図１に示した実施例の変形例を示し
たもので、この実施例では、図１に示したニューラルネ
ットワーク部４４を、線形予測分析部４２からの狭帯域
予測係数より低域部予測係数及び高域部予測係数をそれ
ぞれ推定する第１及び第２のニューラルネットワーク部
４４ａ及び４４ｂで構成しており、合成フィルタ４６
を、ニューラルネットワーク部４４ａ及び４４ｂからの
低域部予測係数及び高域部予測係数を係数とし、それぞ
れ非線形処理部４５からの広帯域予測誤差信号を入力信
号とし、各出力をそれぞれ帯域通過フィルタ４８及び４
９に与える第１及び第２の合成フィルタ４６ａ及び４６
ｂで構成している。FIG. 3 shows a modification of the embodiment shown in FIG. 1. In this embodiment, the neural network unit 44 shown in FIG. It is composed of first and second neural network units 44a and 44b for estimating the low band prediction coefficient and the high band prediction coefficient, respectively.
With the low band prediction coefficient and the high band prediction coefficient from the neural network units 44a and 44b as coefficients, the wide band prediction error signal from the non-linear processing unit 45 as an input signal, and each output as a band pass filter 48 and Four
First and second synthesis filters 46a and 46 to
It consists of b.
【００３１】即ち、線形予測分析部４２で求めた狭帯域
予測係数を入力とするニューラルネットワーク部４４ａ
及び４４ｂを用いて低域部（５０−３００Hz）および高
域部（３４００−７０００Hz）の予測係数をそれぞれ推
定する。That is, the neural network unit 44a to which the narrow band prediction coefficient obtained by the linear prediction analysis unit 42 is input.
And 44b are used to estimate the prediction coefficient of the low frequency band (50-300 Hz) and high frequency band (3400-7000 Hz), respectively.
【００３２】そして、逆フィルタ４３で求めた狭帯域予
測残差信号を非線形処理部４５で広帯域予測残差信号を
生成し、この広帯域予測残差信号をニューラルネットワ
ーク部４４ａ及び４４ｂからの低域部の予測係数および
高域部の予測係数をそれぞれ係数とする合成フィルタ４
６ａ及び４６ｂに通すことにより音声の低域周波数成分
および高域周波数成分をそれぞれ生成する。Then, the narrow-band prediction residual signal obtained by the inverse filter 43 is used by the non-linear processing section 45 to generate a wide-band prediction residual signal, and the wide-band prediction residual signal is supplied to the low-frequency section from the neural network sections 44a and 44b. Filter 4 which uses the prediction coefficient of P and the prediction coefficient of the high frequency band as coefficients
The low-frequency component and the high-frequency component of the voice are generated by passing through 6a and 46b, respectively.
【００３３】各合成フィルタ４６ａ及び４６ｂの出力信
号の低域部（５０−３００Hz）および高域部（３４００
−７０００Hz）をそれぞれ帯域通過フィルタ４８及び４
９を通した後に、これらの二つの信号を、帯域通過フィ
ルタ４７を通した中域周波数帯域（３００−３４００H
z）の入力音声信号に加え合わせることで、広帯域（５
０−７０００Hz）音声信号を生成することができる。The low-frequency part (50-300 Hz) and high-frequency part (3400) of the output signals of the synthesis filters 46a and 46b.
-7000 Hz) with band pass filters 48 and 4 respectively
These two signals are passed through the band pass filter 47 and passed through the middle frequency band (300-3400H).
Wide band (5
0-7000 Hz) audio signal can be generated.
【００３４】第１の実施例では、電話帯域のスペクトル
から７０００Hz帯域のスペクトルを直接推定している
が、上記の低域周波数成分および高域周波数成分はオー
バーラップしているため、変換関数の学習の際に実際に
使用しない中域周波数帯域（３００−３４００Hz）も含
めて学習を行うために無駄が生じることになる。In the first embodiment, the spectrum of the 7000 Hz band is directly estimated from the spectrum of the telephone band, but since the above-mentioned low frequency component and high frequency component overlap, learning of the conversion function is performed. In this case, the learning is performed including the middle frequency band (300-3400 Hz) which is not actually used, which causes waste.
【００３５】第２の実施例ではこのようなことがないた
め、学習の効率を上げることができる。Since the second embodiment does not have such a case, the learning efficiency can be improved.
【００３６】[0036]
【発明の効果】以上説明したように本発明に係る音声通
信システムによれば、広帯域化処理部が、狭帯域受話音
声信号を線形予測分析分析して狭帯域予測係数及び狭帯
域予測残差信号を求め、該狭帯域予測係数からニューラ
ルネットワーク部により広帯域予測係数を推定し、該狭
帯域予測残差信号に対して非線形演算を施して広帯域予
測誤差信号を発生させて該広帯域予測係数を用いて予測
合成を行い、この合成信号から低域周波数成分と高域周
波数成分を抽出した後、狭帯域受話音声信号の中域周波
数成分と合成して広帯域音声信号を求めるように構成し
たので、狭帯域音声の予測残差信号に対して非線形処理
時に発生する高調波成分を利用することにより、少ない
演算量で帯域を増加させることなく広帯域音声信号の再
生を行うことができ、受話音声品質の改善が図れる。As described above, according to the voice communication system of the present invention, the wide band processing unit performs the linear prediction analysis analysis of the narrow band received voice signal to perform the narrow band prediction coefficient and the narrow band prediction residual signal. Then, the neural network unit estimates a wideband prediction coefficient from the narrowband prediction coefficient, performs a nonlinear operation on the narrowband prediction residual signal to generate a wideband prediction error signal, and uses the wideband prediction coefficient. Predictive synthesis is performed, low-frequency components and high-frequency components are extracted from this synthesized signal, and then synthesized with the mid-frequency components of the narrowband received speech signal to obtain a wideband speech signal. By using the harmonic components generated during nonlinear processing for the prediction residual signal of the voice, it is possible to reproduce the wideband voice signal without increasing the bandwidth with a small amount of calculation. , Thereby the improvement of the reception voice quality.
【図１】本発明に係る音声通信システムに用いる広帯域
化処理部の実施例（１）を示したブロック図である。FIG. 1 is a block diagram showing an embodiment (1) of a wide band processing unit used in a voice communication system according to the present invention.
【図２】本発明に係る音声通信システムに用いる広帯域
化処理部におけるニューラルネットワーク部の構成例を
示した図である。FIG. 2 is a diagram showing a configuration example of a neural network unit in a broadband processing unit used in the voice communication system according to the present invention.
【図３】本発明に係る音声通信システムに用いる広帯域
化処理部の実施例（２）を示したブロック図である。FIG. 3 is a block diagram showing an embodiment (2) of the band widening processing unit used in the voice communication system according to the present invention.
【図４】従来から一般的な狭帯域音声通信システムの概
念構成例を示したブロック図である。FIG. 4 is a block diagram showing a conceptual configuration example of a conventional general narrowband voice communication system.
【図５】従来及び本発明に係る狭帯域音声通信システム
に共通な概念構成を示したブロック図である。FIG. 5 is a block diagram showing a conceptual configuration common to conventional narrow band voice communication systems and the present invention.
３ 狭帯域音声受信器 ４ 広帯域化処理部 ４１ Ａ／Ｄ変換器 ４２ 線形予測分析部 ４３ 逆フィルタ ４４，４４ａ，４４ｂ ニューラルネットワーク部 ４５ 非線形処理部 ４６，４６ａ，４６ｂ 合成フィルタ ４７〜４９ 帯域通過フィルタ ５０ 合成部 図中、同一符号は同一又は相当部分を示す。 3 Narrow band voice receiver 4 Broad band processing part 41 A / D converter 42 Linear prediction analysis part 43 Inverse filter 44, 44a, 44b Neural network part 45 Non-linear processing part 46, 46a, 46b Synthesis filter 47-49 Band pass filter 50 Combiner In the drawings, the same reference numerals indicate the same or corresponding parts.
Claims (3)
理部で広帯域化して出力する音声通信システムにおい
て、 該広帯域化処理部が、狭帯域受話音声信号をアナログ／
デジタル変換する変換器と、該変換器の出力信号に対し
て線形予測分析分析を行うことにより狭帯域予測係数を
求める線形予測分析部と、該変換器の出力信号及び該狭
帯域予測係数から狭帯域予測残差信号を求める逆フィル
タと、該狭帯域予測係数から広帯域予測係数を推定する
ニューラルネットワーク部と、該狭帯域予測残差信号に
対して非線形演算を施して広帯域予測誤差信号を発生さ
せる非線形処理部と、該広帯域予測係数を係数とし、該
広帯域予測誤差信号を入力信号とする合成フィルタと、
該変換器の出力信号の第１の周波数帯域を通過させる第
１の帯域通過フィルタと、該合成フィルタの出力信号の
第２及び第３の周波数帯域をそれぞれ通過させる第２及
び第３の帯域通過フィルタと、該第１乃至第３の帯域通
過フィルタの出力信号を入力して広帯域音声信号を合成
する合成部と、を備えていることを特徴とした音声通信
システム。1. A voice communication system in which a narrow band voice signal from a transmission line is widened by a wide band processing unit and output, wherein the wide band processing unit outputs an analog / narrowband received voice signal.
A converter that performs digital conversion, a linear prediction analysis unit that obtains a narrow band prediction coefficient by performing a linear prediction analysis analysis on an output signal of the converter, and a narrow prediction from the output signal of the converter and the narrow band prediction coefficient. An inverse filter for obtaining a band prediction residual signal, a neural network unit for estimating a wide band prediction coefficient from the narrow band prediction coefficient, and a non-linear operation for the narrow band prediction residual signal to generate a wide band prediction error signal. A non-linear processing unit, a synthesis filter having the wideband prediction coefficient as a coefficient, and the wideband prediction error signal as an input signal,
A first bandpass filter for passing a first frequency band of the output signal of the converter, and second and third bandpass for passing a second and third frequency band of the output signal of the synthesis filter, respectively. A voice communication system comprising: a filter; and a synthesizer for inputting output signals of the first to third band pass filters to synthesize a wideband voice signal.
て、該ニューラルネットワーク部が、該狭帯域予測係数
から低域部予測係数及び高域部予測係数をそれぞれ推定
する第１及び第２のニューラルネットワーク部で構成さ
れており、該合成フィルタが、該低域部予測係数及び高
域部予測係数をそれぞれ係数とし、該広帯域予測誤差信
号を入力信号とする第１及び第２の合成フィルタで構成
されていることを特徴とした音声通信システム。2. The voice communication system according to claim 1, wherein the neural network unit estimates a low band prediction coefficient and a high band prediction coefficient from the narrow band prediction coefficient, respectively. The network filter is composed of first and second synthesis filters in which the synthesis filter has the low band prediction coefficient and the high band prediction coefficient as coefficients and the wide band prediction error signal as an input signal. A voice communication system characterized by being provided.
において、該非線形処理部が、全波整流、半波整流、又
は二乗演算を用いることを特徴とした音声通信システ
ム。3. The voice communication system according to claim 1, wherein the non-linear processing section uses full-wave rectification, half-wave rectification, or square operation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP7080034A JPH08278800A (en) | 1995-04-05 | 1995-04-05 | Voice communication system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP7080034A JPH08278800A (en) | 1995-04-05 | 1995-04-05 | Voice communication system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| JPH08278800A true JPH08278800A (en) | 1996-10-22 |
Family
ID=13706986
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| JP7080034A Withdrawn JPH08278800A (en) | 1995-04-05 | 1995-04-05 | Voice communication system |
Country Status (1)
| Country | Link |
|---|---|
| JP (1) | JPH08278800A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2004025625A1 (en) * | 2002-09-12 | 2004-03-25 | Sony Corporation | Signal processing system, signal processing apparatus and method, recording medium, and program |
| JP2006085176A (en) * | 2004-09-17 | 2006-03-30 | Harman Becker Automotive Systems Gmbh | Band enlargement of band-limited audio signal |
| KR100598614B1 (en) * | 2004-08-23 | 2006-07-07 | 에스케이 텔레콤주식회사 | The system and method for wideband expansion of vocal signal using perceptual weighting filter |
| WO2009056027A1 (en) * | 2007-11-02 | 2009-05-07 | Huawei Technologies Co., Ltd. | An audio decoding method and device |
| US8050142B2 (en) | 2007-12-06 | 2011-11-01 | Sanyo Electric Co., Ltd. | Sound collection environment deciding device, sound processing device, electronic appliance, sound collection environment deciding method and sound processing method |
| JP4859670B2 (en) * | 2004-10-27 | 2012-01-25 | パナソニック株式会社 | Speech coding apparatus and speech coding method |
| CN112863477A (en) * | 2020-12-31 | 2021-05-28 | 出门问问(苏州)信息科技有限公司 | Speech synthesis method, device and storage medium |
| CN113345406A (en) * | 2021-05-19 | 2021-09-03 | 苏州奇梦者网络科技有限公司 | Method, apparatus, device and medium for speech synthesis of neural network vocoder |
| JP2022527810A (en) * | 2019-09-18 | 2022-06-06 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Frequency band expansion methods, devices, electronic devices and computer programs |
-
1995
- 1995-04-05 JP JP7080034A patent/JPH08278800A/en not_active Withdrawn
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1302457C (en) * | 2002-09-12 | 2007-02-28 | 索尼株式会社 | Signal processing system, signal processing apparatus and method, recording medium, and program |
| WO2004025625A1 (en) * | 2002-09-12 | 2004-03-25 | Sony Corporation | Signal processing system, signal processing apparatus and method, recording medium, and program |
| US7668319B2 (en) | 2002-09-12 | 2010-02-23 | Sony Corporation | Signal processing system, signal processing apparatus and method, recording medium, and program |
| US7986797B2 (en) | 2002-09-12 | 2011-07-26 | Sony Corporation | Signal processing system, signal processing apparatus and method, recording medium, and program |
| KR100598614B1 (en) * | 2004-08-23 | 2006-07-07 | 에스케이 텔레콤주식회사 | The system and method for wideband expansion of vocal signal using perceptual weighting filter |
| JP2006085176A (en) * | 2004-09-17 | 2006-03-30 | Harman Becker Automotive Systems Gmbh | Band enlargement of band-limited audio signal |
| JP4859670B2 (en) * | 2004-10-27 | 2012-01-25 | パナソニック株式会社 | Speech coding apparatus and speech coding method |
| WO2009056027A1 (en) * | 2007-11-02 | 2009-05-07 | Huawei Technologies Co., Ltd. | An audio decoding method and device |
| US8473301B2 (en) | 2007-11-02 | 2013-06-25 | Huawei Technologies Co., Ltd. | Method and apparatus for audio decoding |
| US8050142B2 (en) | 2007-12-06 | 2011-11-01 | Sanyo Electric Co., Ltd. | Sound collection environment deciding device, sound processing device, electronic appliance, sound collection environment deciding method and sound processing method |
| JP2022527810A (en) * | 2019-09-18 | 2022-06-06 | ▲騰▼▲訊▼科技（深▲セン▼）有限公司 | Frequency band expansion methods, devices, electronic devices and computer programs |
| US12002479B2 (en) | 2019-09-18 | 2024-06-04 | Tencent Technology (Shenzhen) Company Limited | Bandwidth extension method and apparatus, electronic device, and computer-readable storage medium |
| CN112863477A (en) * | 2020-12-31 | 2021-05-28 | 出门问问(苏州)信息科技有限公司 | Speech synthesis method, device and storage medium |
| CN112863477B (en) * | 2020-12-31 | 2023-06-27 | 出门问问(苏州)信息科技有限公司 | Speech synthesis method, device and storage medium |
| CN113345406A (en) * | 2021-05-19 | 2021-09-03 | 苏州奇梦者网络科技有限公司 | Method, apparatus, device and medium for speech synthesis of neural network vocoder |
| CN113345406B (en) * | 2021-05-19 | 2024-01-09 | 苏州奇梦者网络科技有限公司 | Method, device, equipment and medium for synthesizing voice of neural network vocoder |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jayant et al. | Signal compression based on models of human perception | |
| US5701346A (en) | Method of coding a plurality of audio signals | |
| KR100299528B1 (en) | Apparatus and method for encoding / decoding audio signal using intensity-stereo process and prediction process | |
| US6496795B1 (en) | Modulated complex lapped transform for integrated signal enhancement and coding | |
| CN102016983B (en) | Apparatus for mixing plurality of input data streams | |
| JP3283413B2 (en) | Encoding / decoding method, encoding device and decoding device | |
| KR100986150B1 (en) | Audio coding system using characteristics of a decoded signal to adapt synthesized spectral components | |
| JP4223679B2 (en) | Low bit rate multiplex audio channel encoding / decoding method and apparatus | |
| JP3513292B2 (en) | Noise weight filtering method | |
| US5383184A (en) | Multi-speaker conferencing over narrowband channels | |
| WO2004023457A1 (en) | Sound encoding apparatus and sound encoding method | |
| CN105895107A (en) | Audio packet loss concealment by transform interpolation | |
| AU2003243441B2 (en) | Audio coding system using characteristics of a decoded signal to adapt synthesized spectral components | |
| JP2000172300A (en) | Method for generating wide band signal based on narrow band signal, device for realizing such method and telephone system equipment containing such device | |
| JP3219762B2 (en) | Signal transmission method | |
| US9118805B2 (en) | Multi-point connection device, signal analysis and device, method, and program | |
| JPH09204200A (en) | Conferencing system | |
| JPH08278800A (en) | Voice communication system | |
| KR20020081388A (en) | Speech decoder and a method for decoding speech | |
| US20110002225A1 (en) | Signal analysis/control system and method, signal control apparatus and method, and program | |
| KR100952065B1 (en) | Coding method, apparatus, decoding method, and apparatus | |
| JPH10340097A (en) | Comfortable noise generator, voice encoder including its component and decoder | |
| EP0795755A2 (en) | Method of non-harmonic analysis of waveforms for synthesis, interpolation and extrapolation | |
| JPH07147566A (en) | Sound signal transmitter | |
| CN101689372B (en) | Signal analysis device, signal control device, its system, method, and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| A300 | Application deemed to be withdrawn because no request for examination was validly filed |
Free format text: JAPANESE INTERMEDIATE CODE: A300 Effective date: 20020702 |