WO2022154341A1 - 음성 합성 시스템의 동작방법 - Google Patents
음성 합성 시스템의 동작방법 Download PDFInfo
- Publication number
- WO2022154341A1 WO2022154341A1 PCT/KR2021/095116 KR2021095116W WO2022154341A1 WO 2022154341 A1 WO2022154341 A1 WO 2022154341A1 KR 2021095116 W KR2021095116 W KR 2021095116W WO 2022154341 A1 WO2022154341 A1 WO 2022154341A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- text
- combined
- speech synthesis
- speech
- voice
- 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.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/08—Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/02—Methods for producing synthetic speech; Speech synthesisers
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/02—Methods for producing synthetic speech; Speech synthesisers
- G10L13/04—Details of speech synthesis systems, e.g. synthesiser structure or memory management
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L13/00—Speech synthesis; Text to speech systems
- G10L13/06—Elementary speech units used in speech synthesisers; Concatenation rules
- G10L13/07—Concatenation rules
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
- G10L25/18—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being spectral information of each sub-band
Definitions
- the present invention relates to a method of operating a speech synthesis system, and more particularly, to a method of operating a speech synthesis system capable of outputting a target synthesized voice corresponding to a long target text.
- AI Artificial intelligence
- STT speech recognition
- NLU natural language understanding
- NGL natural language generation
- TTS speech synthesis
- Speech synthesis technology is a key technology for realizing an interactive user interface through artificial intelligence.
- Existing speech synthesis is a method of creating waveforms by combining words, syllables, and phonemes, which are fixed length units (1st generation), and a method of connecting variable synthesis units using a corpus (2nd generation) to a 3rd generation model. developed.
- the 3rd generation model implemented high-quality speech synthesis using a database of an appropriate size by applying the HMM (Hidden Markov Model) method, which is mainly used for acoustic modeling for speech recognition, to speech synthesis.
- HMM Hidden Markov Model
- the existing voice synthesis required at least 5 hours of the speaker's voice data and 10 hours or more for high-quality voice output.
- a method of operating a speech synthesis system comprises the steps of: inputting a first text, a first voice for the first text, a second text, and a second voice for the second text, the first and second texts and generating a voice shout model learned by applying the first and second voices to curriculum learning.
- a target text When a target text is input for voice output, a target corresponding to the target text based on the speech synthesis model outputting a synthesized voice, wherein the generating of the voice synthesis model includes: generating a combined text combining the first and second texts and a combined voice combining the first and second voices; and the combined text and
- the method may include adding the combined text and the combined speech to the speech synthesis model when an error rate when learning and combining the combined speech is smaller than a set reference rate.
- the combined text may include the first and second texts and a text token for discriminating the first and second texts.
- the combined voice may include the first and second voices and a mel spectrogram-token for discriminating the first and second voices.
- the text token and the melspectrogram token may have a time interval of 1 second to 2 seconds.
- the text token and the melspectrogram token may be a silent section.
- the adding to the speech synthesis model may include combining the text token and the melspectrogram token as a reference.
- the method Prior to the step of adding to the speech synthesis model, the method further includes initializing the combined text and the combined speech when a batch size is smaller than a set reference batch size when learning and combining the combined text and the combined speech. can do.
- the adding to the speech synthesis model may include initializing the combined text and the combined speech when the error rate is greater than the reference rate.
- the operation method of the speech synthesis system according to the present invention has an advantage in that it is easy to synthesize speech for long texts by generating a speech synthesis model from a short text to a long text by learning through curriculum learning.
- the operating method of the speech synthesis system according to the present invention has the advantage of outputting a natural voice when synthesizing a long text.
- FIG. 1 is a control block diagram showing a control configuration of a speech synthesis system according to the present invention.
- Figure 2 is a simplified diagram for explaining the combined text and combined voice according to the present invention.
- 3 and 4 are diagrams and tables showing experimental results for the speech synthesis system according to the present invention.
- FIG. 5 is a flowchart illustrating a method of operating a speech synthesis system according to the present invention.
- first, second, A, and B may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.
- FIG. 1 is a control block diagram showing a control configuration of a speech synthesis system according to the present invention
- FIG. 2 is a simplified diagram for explaining a combined text and a combined voice according to the present invention.
- the speech synthesis system 100 may include an encoder 110 , an attention 120 , and a decoder 130 .
- the encoder 110 may process the input text, and the decoder 130 may output a voice.
- the encoder 110 may compress the text into a vector, and the decoder 130 may output the voice output from the attention 120 .
- the voice may be a Mel spectrogram (Me1 spectrogram).
- the encoder 110 may convert the text into a number vector representing text information, but is not limited thereto.
- the attention 120 may generate a voice from text.
- the attention 120 may minimize the gradient loss phenomenon of the input sequence by utilizing the attention mechanism.
- the attention mechanism is expressed in the form of a function as follows.
- Q means Query, which is a value that reflects the hidden state in the decoder 130 at the time.
- K stands for Key
- V stands for Value, which reflects the hidden states of the encoder 110 at all times.
- the attention function emphasizes the important part by calculating the similarity of the query for the key and reflecting it in the value.
- the decoder 130 dot product the query reflecting the value of the current time and the key reflecting the encoder value, and then take the softmax, it is expressed in the form of a weight to determine which part of the encoder 110 the query should focus on to help the prediction. .
- each weight is called an attention weight.
- an attention value is obtained. Since the attention value includes the context of the encoder 110 , it is also called a context vector.
- the output sequence is obtained by concatenating the context vector with the hidden state of the current time of the decoder 130 .
- the attention 130 may generate a voice synthesis model by synthesizing text and voice.
- the attention 130 when the learning data is input, the attention 130 generates a learned speech synthesis model by applying the training data to a set curriculum learning, and when the target text for voice output is input, the speech synthesis model is based on the to generate a target synthesized voice corresponding to the target text.
- the learning data is described as consisting of a first text and a first voice for the first text, and a second voice for the second text and the second text, but limited to the number of texts and the number of voices do not put
- the attention 130 may receive a vector for the first text from the encoder 110 and a melspectrogram for the first voice from the decoder 120 . .
- the attention 130 receives the vector for the second text from the encoder 110 and receives the mel spectrogram for the second voice from the decoder 120 .
- the attention 130 may learn by applying the first and second texts and the first and second voices to a set curriculum learning.
- the attention 130 may individually learn the first and second texts and the first and second voices.
- the attention 130 may generate a combined text combining the first and second texts and a combined voice combining the first and second voices.
- the combined text may include a text token for distinguishing the first and second texts and the first and second texts
- the combined voice is for distinguishing the first and second voices and the first and second voices. It may include a mel spectrogram-token.
- the text token and the melspectrogram token are a time interval of 1 second to 2 seconds, and may be a silence interval.
- the text token and the melspectrogram token are tools for allowing two first and second texts to be naturally connected to each other and recognized as one text.
- the attention 130 has been described as generating and learning one combined text and one combined voice from two first and second texts and two first and second voices, but there are a plurality of texts and a plurality of voices. In this case, a plurality of combined text and combined voice may be generated, but the present invention is not limited thereto.
- the attention 130 may initialize the combined text and the combined voice when the batch size is smaller than the set reference batch size when the combined text and the combined voice are learned and combined.
- the attention 130 may recombine the first and second texts and the first and second voices, or may not generate the combined text and the combined voice.
- the attention 130 may add the combined text and the combined speech to the speech synthesis model when an error rate is smaller than a set reference rate when learning and combining the combined text and the combined speech.
- the attention 130 may initialize the combined text and the combined voice.
- the attention 130 may output a target synthesized voice corresponding to the target text based on the voice synthesis model.
- the attention 130 may output the target synthesized voice to the decoder 120 .
- 3 and 4 are diagrams and tables showing experimental results for the speech synthesis system according to the present invention.
- the speech synthesis system 100 may perform speech synthesis by using Tacotron2 as a basic model and additionally applying a model for curriculum learning.
- the batch size was basically 12, and when learning by combining n sentences to synthesize long sentences through curriculum learning within the limited GPU capacity, the batch size is automatically set to 1 It is set to shrink to a size of /n.
- the speech synthesis system 100 has the advantage of synthesizing long sentences at once.
- the speech synthesis system 100 was subjected to a speech synthesis test using a script from a novel Harry Potter book, and evaluation was performed according to the length and time of the synthesized speech.
- Figure 3 compares the proposed model using the Tacotron model using content based attention and the Tacotron2 model using location sensitive attention to compare the length of sentences that can be synthesized with the existing model.
- the model we proposed does not exceed the CER of 10% until synthesizing a voice (about 4400 characters) with a length of 5 minutes and 30 seconds, whereas the content based attention model is 10 seconds and the location sensitive attention model is 30 seconds In the invitation, it was confirmed that over 20%.
- DLTTS1 curriculum learning is not applied
- DLTTS2 when only two sentences are applied
- DLTTS3 curriculum learning ( curriculum learning) was applied
- the model provided by the speech synthesis system 100 of the present invention exhibits a very low attention error rate at the document-level compared to the existing model.
- the document-level neural TTS model showed a relatively low attention error rate even when synthesizing sentences longer than 5 minutes. This shows that document-level sentences can also be synthesized stably.
- the attention error rate exceeded 50% in a sentence environment of 5 minutes or more, whereas when curriculum learning was executed with 2 sentences, 25%, 3 When executed as a sentence, it was found that an attention error rate of about 1% was measured. Through this, we found that curriculum learning is an essential element when synthesizing speech in units of documents.
- FIG. 5 is a flowchart illustrating a method of operating a speech synthesis system according to the present invention.
- the speech synthesis system 100 may input a first text, a first voice for the first text, a second text, and a second voice for the second text ( S110 ).
- the speech synthesis system 100 applies the first and second texts and the first and second voices to curriculum learning to combine the first and second texts and combine the first and second voices.
- a voice may be generated (S120).
- the speech synthesis system 100 determines whether a batch size is smaller than a set reference batch size when learning and combining the combined text and the combined speech (S130), and if smaller than the reference batch size, the combined text and the combined The voice may be initialized (S140).
- the speech synthesis system 100 determines whether an error rate when learning and combining the combined text and the combined speech is smaller than a set reference rate (S150), and the reference rate If less than, the combined text and the combined voice may be generated and added to the speech synthesis model (S160).
- the speech synthesis system 100 may initialize the combined text and the combined voice (S170).
- the speech synthesis system 100 may output a target synthesized voice corresponding to the target text based on the speech synthesis model (S180).
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Signal Processing (AREA)
- Machine Translation (AREA)
- Electrically Operated Instructional Devices (AREA)
Abstract
Description
Claims (8)
- 제1 텍스트와 상기 제1 텍스트에 대한 제1 음성 및 제2 텍스트와 상기 제2 텍스트에 대한 제2 음성이 입력되는 단계;상기 제1, 2 텍스트 및 상기 제1, 2 음성을 커리큘럼 러닝(Curriculum learning)에 적용하여 학습한 음성 함성 모델을 생성하는 단계; 및음성 출력을 위한 대상 텍스트 입력 시, 상기 음성 합성 모델을 기반으로 상기 대상 텍스트에 대응하는 대상 합성 음성을 출력하는 단계를 포함하고,상기 음성 합성 모델을 생성하는 단계는,상기 제1, 2 텍스트를 결합한 결합 텍스트 및 상기 제1, 2 음성을 결합한 결합 음성을 생성하는 단계; 및상기 결합 텍스트 및 상기 결합 음성의 학습 결합 시 에러 레이트(error rate)가 설정된 기준 레이트(reference rate)보다 작으면 상기 결합 텍스트 및 상기 결합 음성을 상기 음성 합성 모델에 추가하는 단계를 포함하는,음성 합성 시스템의 동작방법.
- 제 1 항에 있어서,상기 결합 텍스트는,상기 제1, 2 텍스트 및 상기 제1, 2 텍스트를 구분하는 텍스트 토큰(text token)을 포함하는,음성 합성 시스템의 동작방법.
- 제 2 항에 있어서,상기 결합 음성은,상기 제1, 2 음성 및 상기 제1, 2 음성을 구분하는 멜스펙트로그램 토큰(mel spectrogram-token)을 포함하는,음성 합성 시스템의 동작방법.
- 제 3 항에 있어서,상기 텍스트 토큰 및 상기 멜스펙트로그램 토큰은,1초 내지 2초의 시간 구간을 갖는,음성 합성 시스템의 동작방법.
- 제 3 항에 있어서,상기 텍스트 토큰 및 상기 멜스펙트로그램 토큰은,묵음 구간인,음성 합성 시스템의 동작방법.
- 제 3 항에 있어서,상기 음성 합성 모델에 추가하는 단계는,상기 텍스트 토큰 및 상기 멜스펙트로그램 토큰을 기준으로 결합하는,음성 합성 시스템의 동작방법.
- 제 1 항에 있어서,상기 음성 합성 모델에 추가하는 단계 이전에,상기 결합 텍스트 및 상기 결합 음성의 학습 결합 시 배치 사이즈(batch size)가 설정된 기준 배치 사이즈보다 작으면 상기 결합 텍스트 및 상기 결합 음성을 초기화하는 단계를 더 포함하는,음성 합성 시스템의 동작방법.
- 제 1 항에 있어서,상기 음성 합성 모델에 추가하는 단계는,상기 에러 레이트가 상기 기준 레이트보다 크면 상기 결합 텍스트 및 상기 결합 음성을 초기화하는,음성 합성 시스템의 동작방법.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US18/271,933 US12431118B2 (en) | 2021-01-13 | 2021-12-02 | Operation method of speech synthesis system |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| KR10-2021-0004856 | 2021-01-13 | ||
| KR1020210004856A KR20220102476A (ko) | 2021-01-13 | 2021-01-13 | 음성 합성 시스템의 동작방법 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2022154341A1 true WO2022154341A1 (ko) | 2022-07-21 |
Family
ID=82448364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/KR2021/095116 Ceased WO2022154341A1 (ko) | 2021-01-13 | 2021-12-02 | 음성 합성 시스템의 동작방법 |
Country Status (3)
| Country | Link |
|---|---|
| US (1) | US12431118B2 (ko) |
| KR (2) | KR20220102476A (ko) |
| WO (1) | WO2022154341A1 (ko) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010237323A (ja) * | 2009-03-30 | 2010-10-21 | Toshiba Corp | 音声モデル生成装置、音声合成装置、音声モデル生成プログラム、音声合成プログラム、音声モデル生成方法および音声合成方法 |
| KR20190085882A (ko) * | 2018-01-11 | 2019-07-19 | 네오사피엔스 주식회사 | 기계학습을 이용한 텍스트-음성 합성 방법, 장치 및 컴퓨터 판독가능한 저장매체 |
| JP2020126141A (ja) * | 2019-02-05 | 2020-08-20 | 日本電信電話株式会社 | 音響モデル学習装置、音響モデル学習方法、プログラム |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10319374B2 (en) | 2015-11-25 | 2019-06-11 | Baidu USA, LLC | Deployed end-to-end speech recognition |
| WO2018167522A1 (en) * | 2017-03-14 | 2018-09-20 | Google Llc | Speech synthesis unit selection |
-
2021
- 2021-01-13 KR KR1020210004856A patent/KR20220102476A/ko not_active Ceased
- 2021-12-02 WO PCT/KR2021/095116 patent/WO2022154341A1/ko not_active Ceased
- 2021-12-02 US US18/271,933 patent/US12431118B2/en active Active
-
2023
- 2023-05-11 KR KR1020230060960A patent/KR102649028B1/ko active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2010237323A (ja) * | 2009-03-30 | 2010-10-21 | Toshiba Corp | 音声モデル生成装置、音声合成装置、音声モデル生成プログラム、音声合成プログラム、音声モデル生成方法および音声合成方法 |
| KR20190085882A (ko) * | 2018-01-11 | 2019-07-19 | 네오사피엔스 주식회사 | 기계학습을 이용한 텍스트-음성 합성 방법, 장치 및 컴퓨터 판독가능한 저장매체 |
| JP2020126141A (ja) * | 2019-02-05 | 2020-08-20 | 日本電信電話株式会社 | 音響モデル学習装置、音響モデル学習方法、プログラム |
Non-Patent Citations (2)
| Title |
|---|
| HWANG SUNG-WOONG; CHANG JOON-HYUK: "Document-Level Neural TTS Using Curriculum Learning and Attention Masking", IEEE ACCESS, vol. 9, 4 January 2021 (2021-01-04), IEEE, USA, pages 8954 - 8960, XP011831686, DOI: 10.1109/ACCESS.2020.3049073 * |
| YAHUAN CONG; RAN ZHANG; JIAN LUAN: "PPSpeech: Phrase based Parallel End-to-End TTS System", ARXIV.ORG, CORNELL UNIVERSITY LIBRARY, ITHACA, NY 14853, 6 August 2020 (2020-08-06), Ithaca, NY 14853, XP081734812 * |
Also Published As
| Publication number | Publication date |
|---|---|
| US12431118B2 (en) | 2025-09-30 |
| US20240153486A1 (en) | 2024-05-09 |
| KR20230070423A (ko) | 2023-05-23 |
| KR20220102476A (ko) | 2022-07-20 |
| KR102649028B1 (ko) | 2024-03-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shah et al. | What all do audio transformer models hear? probing acoustic representations for language delivery and its structure | |
| KR102581346B1 (ko) | 다국어 음성 합성 및 언어간 음성 복제 | |
| CN115547293B (zh) | 一种基于分层韵律预测的多语言语音合成方法及系统 | |
| US20210256961A1 (en) | Method and System for Parametric Speech Synthesis | |
| CN112352275A (zh) | 具有多级别文本信息的神经文本到语音合成 | |
| CN115424604B (zh) | 一种基于对抗生成网络的语音合成模型的训练方法 | |
| KR970707529A (ko) | 음성 인식 장치 및 음성 인식 장치용 어휘 발생 방법(speech recognition) | |
| WO2021162362A1 (ko) | 음성 인식 모델을 학습시키는 방법 및 상기 방법을 이용하여 학습된 음성 인식 장치 | |
| Cao et al. | Code-switched speech synthesis using bilingual phonetic posteriorgram with only monolingual corpora | |
| CN113823259A (zh) | 将文本数据转换为音素序列的方法及设备 | |
| WO2022203152A1 (ko) | 다화자 훈련 데이터셋에 기초한 음성합성 방법 및 장치 | |
| Qian et al. | Optimizing data usage for low-resource speech recognition | |
| CN118193702A (zh) | 用于英语教学的智能化人机交互系统及方法 | |
| Mortensen et al. | AlloVera: A multilingual allophone database | |
| WO2019098589A1 (ko) | 자동 음성인식 장치 및 방법 | |
| Love | Integrating languages | |
| WO2023229117A1 (ko) | 대화형 가상 아바타의 구현 방법 | |
| Lee et al. | Embedding Articulatory Constraints for Low-resource Speech Recognition Based on Large Pre-trained Model. | |
| WO2022065603A1 (ko) | 텍스트 기반의 음성 합성 방법 및 장치 | |
| CN114420089B (zh) | 语音合成方法、装置和计算机可读存储介质 | |
| CN109859746B (zh) | 一种基于tts的语音识别语料库生成方法及系统 | |
| CN114566143B (zh) | 一种可局部修改内容的语音合成方法及语音合成系统 | |
| KR102649028B1 (ko) | 음성 합성 시스템의 동작방법 | |
| Magoshi et al. | Multi-lingual and Zero-Shot Speech Recognition by Incorporating Classification of Language-Independent Articulatory Features | |
| Ainsworth et al. | Connectionist architectures for a text-to-speech system. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21919963 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 18271933 Country of ref document: US |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 21919963 Country of ref document: EP Kind code of ref document: A1 |
|
| WWG | Wipo information: grant in national office |
Ref document number: 18271933 Country of ref document: US |