A Hausa speech synthesis method and system

Abstract

The invention discloses a Hausa speech synthesis method and system. After the input Hausa text is processed for polyphone disambiguation, it is input into a Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech; the spectral features are input into a Hausa vocoder to map the spectral features into speech. The Hausa text input by the user is processed at the front end, and the contextual semantics of the language are used to measure the connection between word tones using a conditional random field. The entire recognition model has a simple structure and a small number of model parameters. It is easy to deploy on a cloud or mobile device. The recognition result can be quickly and accurately calculated for the input Hausa text, and the accuracy of subsequent synthesis is improved. For different end-to-end and mobile application scenarios, different combinations of acoustic models and vocoders are selected to improve the accuracy, naturalness and synthesis efficiency of Hausa speech synthesis, and can be applied to various speech interaction scenarios in the Hausa context.

Description

Technical Field

[0001] The present invention relates to the field of speech synthesis, and in particular to a Hausa speech synthesis method and system. Background Art

[0002] As a carrier of human information transmission, speech is inseparable from people's lives. With the rapid development of mobile Internet and speech processing technology, speech plays an increasingly important role in our lives. Speech synthesis technology is one of the core technologies in speech processing technology. It aims to convert any text into speech and is indispensable in human-computer interaction. The speech synthesis system consists of two modules: the front-end and the back-end. The front-end module usually analyzes the text to obtain the pronunciation and rhythm of the text. For speech with tones, such as Chinese and most African languages, the front-end processing module also needs to obtain the tone of the text. The back-end module mainly models the sound, obtains a series of acoustic features of the text, and further obtains the waveform of the sound.

[0003] Hausa is one of the three major languages ​​in Africa and is widely used in West Africa. Hausa is also a tonal language. Different tones contain different semantics and pronunciations, but the tones are not clearly marked in formal writing. In Hausa, each vowel letter (a, e, i, o, u) may have three tones: low tone, high tone or falling tone. In addition, the length of the vowel pronunciation also indicates different semantics and pronunciations. Therefore, a Hausa word may contain multiple tones. Hausa speech synthesis refers to converting Hausa text into corresponding speech. Therefore, the problem that must be solved is to determine the tone of each word in the Hausa text. However, the tone is often related to the contextual semantics of the text. A word may have different tones under different semantics. It is difficult to cover all situations based on handwritten rules to accurately determine the tone of each word, which makes Hausa tone recognition a very challenging task.

[0004] In recent years, with the development of deep learning, more and more speech synthesis methods have been developed. In addition to English, speech synthesis systems for different languages ​​such as Chinese, English, Japanese, and Korean have been widely used. Hausa is an important language in Africa with a large number of speakers, but there is still a lack of speech synthesis methods and systems for Hausa. Summary of the invention

[0005] Therefore, in order to overcome the defect of the prior art that there is a lack of a speech synthesis method and system for Hausa, the present invention provides a Hausa speech synthesis method and system.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a Hausa speech synthesis method, comprising: performing polyphonetic word disambiguation processing on an input Hausa text;

[0008] Input the processed Hausa text into the Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech;

[0009] The spectral features are input into a Hausa vocoder to map the spectral features into speech.

[0010] Preferably, the process of performing polyphone disambiguation processing on a Hausa text comprises:

[0011] The Hausa text is input into a deep neural network to output the tone score of each word, and the optimal tone recognition result is obtained through a decoding module. The deep neural network includes: a word vector module, a context semantic extraction module, a fully connected layer, and a conditional random field module, wherein:

[0012] The word vector module is used to obtain the pre-trained word vector of each word in the Hausa text to be recognized. The obtained word vector is sent to the context semantic extraction module to obtain the semantic representation of each word containing context information. After obtaining the representation of each word with context semantic information, the representation of each word is sent to the fully connected layer to obtain the score on each tone. After obtaining the transfer score between tones through the conditional random field, combined with the previously calculated scores of each word on each tone, the decoding module is used to decode the optimal tone recognition result for the entire text.

[0013] Preferably, the context semantic extraction module consists of a one-layer bidirectional long short-term memory network.

[0014] Preferably, the decoding module uses a Viterbi algorithm to calculate the tone sequence with the maximum score of the Hausa text.

[0015] Preferably, the Hausa acoustic module model adopts Tacotron2, a sequence-to-sequence generative model with an attention mechanism, including three sub-modules: an encoder, a decoder, and a post-processing network; wherein the encoder module processes the input Hausa text into text features with contextual information, the decoder is responsible for mapping these text features into corresponding spectral features, and the post-processing network plays a role in supplementing the details in the spectral features; the Hausa vocoder adopts WaveNet or MelGAN.

[0016] Preferably, the Hausa acoustic model uses Tacotron 2 compressed according to a preset compression ratio, and the Hausa vocoder selects Griffin-Lim or a neural network vocoder Squeeze_wave.

[0017] Preferably, the Tacotron2 includes an encoder, a decoder and an attention module, wherein the encoder includes: 1 embedding layer, 3 convolutional layers, and 1 bidirectional long short-term memory layer; the decoder includes: 2 fully connected layers, 2 unidirectional long short-term memory layers, 1 mapping layer and 5 convolutional layers. The number of filters in the convolutional layers of the encoder and decoder and the number of hidden units in the LSTM and fully connected layers are compressed to 50% of the original as the network structure of the acoustic model.

[0018] In a second aspect, an embodiment of the present invention provides a Hausa speech synthesis system, comprising:

[0019] A Hausa language front-end processing module is used to perform polyphone disambiguation processing and digital processing on the input Hausa language text;

[0020] A Hausa acoustic model module is used to input the processed Hausa text into the Hausa acoustic model to obtain the spectrum features corresponding to the Hausa speech;

[0021] The Hausa vocoder module is used to input the frequency spectrum features into a Hausa vocoder and map the frequency spectrum features into speech.

[0022] In a third aspect, an embodiment of the present invention provides a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the Hausa speech synthesis method described in the first aspect of the embodiment of the present invention.

[0023] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the Hausa speech synthesis method described in the first aspect. .

[0024] The technical solution of the present invention has the following advantages:

[0025] The Hausa speech synthesis method and system provided by the embodiment of the present invention performs polyphone disambiguation processing on the input Hausa text; inputs the processed Hausa text into the Hausa acoustic model to obtain the spectrum features corresponding to the Hausa speech; inputs the spectrum features into the Hausa vocoder to map the spectrum features into speech. The front-end processing is performed on the Hausa text input by the user, and the contextual semantics of the language are used to measure the connection between word tones using conditional random fields. The entire recognition model has a simple structure and a small number of model parameters, and is easy to be deployed on cloud or mobile devices. The recognition result can be quickly and accurately calculated for the input Hausa text, and the accuracy of subsequent synthesis is improved. Different combinations of acoustic models and vocoders are selected for different end-to-end and mobile application scenarios, which can meet the corresponding needs, improve the accuracy, naturalness and synthesis efficiency of Hausa speech synthesis, and can be applied to various speech interaction scenarios in the Hausa context. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0027] Figure 1 A flowchart of an example of a Hausa speech synthesis method provided in an embodiment of the present invention;

[0028] Figure 2 A schematic diagram of a random field learning transfer scores between each tone and each tone provided in an embodiment of the present invention;

[0029] Figure 3 A module composition diagram of an example of a Hausa speech synthesis system provided in an embodiment of the present invention;

[0030] Figure 4 A composition diagram of a specific example of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] Example 1

[0034] The embodiment of the present invention provides a Hausa speech synthesis method, which can be applied to end-to-end and mobile Hausa speech synthesis application scenarios, such as Figure 1 As shown, including:

[0035] Step S1: Perform polyphone disambiguation processing on the input Hausa text.

[0036] Specifically, the Hausa text is input into a deep neural network to output the tone score of each word, and the optimal tone recognition result is obtained through a decoding module. The deep neural network includes: a word vector module, a context semantic extraction module, a fully connected layer, and a conditional random field module, wherein:

[0037] The Hausa text to be recognized is used to obtain the pre-trained word vector of each word in the text through the word vector module (the pre-training process of the word vector adopts the current common method, that is, a large amount of Hausa text is sent to the language model such as GPT for training, and finally the semantically rich word vector of each word can be obtained, which is not described in detail here), and the obtained word vector is sent to the context semantic extraction module (composed of a layer of bidirectional long short-term memory network BiLSTM. Compared with the unidirectional LSTM, BiLSTM splices the hidden state obtained by forward and backward calculations, which can capture the forward and backward information at the same time and extract the context semantic information more effectively) to obtain the semantic representation of each word containing context information. After obtaining the representation of each word with context semantic information, the representation of each word is sent to the fully connected layer to obtain the score on each tone. After obtaining the transfer score between the tones through the conditional random field, combined with the previously calculated scores of each word on each tone, the decoding module is used to decode the optimal tone recognition result for the entire text.

[0038] The most direct approach in the prior art is to send the output of the fully connected layer to the softmax layer to obtain the probability of belonging to each tone, and take the corresponding tone with the highest probability as the tone of the word. However, this method does not take into account the influence of the recognition results of the previous and next words when making decisions for each word. In fact, the recognition results of the previous and next words may have a greater impact on the recognition of the current word. For example, the tone of a word in the text is determined to be 1. According to experience, the tone of the word behind it is very likely to be 4, and it is basically impossible to be tone 2. Therefore, in order to introduce this kind of connection between the previous and next words into the method provided in the embodiment of the present invention, a conditional random field module is added after the fully connected layer. The conditional random field measures the transfer relationship between different tones, so that the entire input text has a more accurate tone recognition result as a whole. After obtaining the transfer scores between the tones through the conditional random field, combined with the previously calculated scores of each word on each tone, the Viterbi algorithm is used to decode the optimal tone recognition result for the entire text.

[0039] In a specific embodiment, the conditional random field model learns the transition scores between each tone, such as Figure 2 As shown in the figure, assuming that there are three tones: T1, T2 and T3, the conditional random field model will generate a 3×3 transfer matrix. The elements in the first row and first column of the matrix represent the transfer score of tone T1 to tone T1, and the elements in the i-th row and j-th column represent the transfer score of tone T i Tone T j For a word sequence in a sentence, we hope to obtain a tone sequence that maximizes the sum of the word score and the transfer score. Figure 2 As shown, for the word sequence<w1,w2,w3,w4> , without considering the transfer score, the final predicted tone sequence is<T1,T1,T2,T3> ; After taking the transition relationship between tones into account, the tone sequence obtained is<T1,T2,T2,T3> The sum of the transfer and self scores is 0.8+0.9+0.3+0.8+0.9+0.7+0.7=5.2. If we ignore the transfer relationship and directly feed it into the softmax layer, we get the tone sequence<T1,T1,T2,T3> The total score is 0.8+0.1+0.4+0.8+0.9+0.7+0.7=4.4, which cannot guarantee the global optimum after considering the transfer score. When training the entire network, the real tone sequence is guaranteed to have the highest score to update the network parameters and transfer matrix. When predicting, the word score and transfer score matrix can be obtained. Based on the Viterbi algorithm, the tone sequence with the highest score can be obtained.

[0040] Step S2: Input the processed Hausa text into the Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech.

[0041] Step S3: input the spectrum features into a vocoder to map the spectrum features into speech.

[0042] In a specific embodiment, if the end-to-end Hausa speech synthesis is to be performed, the Hausa acoustic model uses the sequence-to-sequence generation model Tacotron 2 with an attention mechanism. If the Hausa vocoder needs to synthesize high-fidelity speech without considering the delay caused by the system, a vocoder such as WaveNet can be used; otherwise, a vocoder such as MelGAN that can be accelerated by GPU parallel computing can be selected. Tacotron 2 is a sequence-to-sequence generation model with an attention mechanism, which realizes the modeling from text to spectral features. A large number of experiments have proved that this model can learn the alignment between text and spectrum, generate natural and rhythmic speech, and meet the requirements of modern speech synthesis for naturalness. It includes three submodules: encoder, decoder and post-processing network; among them, the encoder module processes the input Hausa text into text features with context information, the decoder is responsible for mapping these text features into corresponding spectral features, and the post-processing network plays a role in supplementing the details in the spectral features.

[0043] In a specific embodiment, if the Hausa speech synthesis is for mobile terminals, the model needs to be compressed due to the limited resources of the mobile terminals. In order to use the tacotron2 module on the mobile terminal device, the embodiment of the present invention cuts the number of channels of the model to varying degrees, so that the amount of calculation and the number of parameters of the model are greatly reduced. In order to improve the reasoning efficiency of autoregression, a strategy of decoding multiple frames each time is adopted to multiply the reasoning speed on the mobile terminal with almost no loss of sound quality. Thanks to the above two strategies, the Tacotron2 model can perform low-latency reasoning on mobile devices.

[0044] The Tacotron2 model includes: encoder, decoder and attention module. The encoder includes: 1 embedding layer, 3 convolution layers, 1 bidirectional long short-term memory layer; the decoder includes: 2 fully connected layers, 2 unidirectional long short-term memory layers, 1 mapping layer and 5 convolution layers. In the actual test process, the number of convolution layer filters of the encoder and decoder parts was reduced to 75%, 50%, and 25% respectively, and the number of hidden units of the LSTM and fully connected layers of the encoder and decoder parts was reduced to 75%, 50%, and 25%. Finally, the experiment concluded that the effect and performance obtained by cutting all to 50% are the most suitable parameters for mobile terminals.

[0045] 100％ 75％ 50％ 25％ Encoder CNN 512 384 256 128 Encoder LSTM 512 384 256 128 Decoder FC 256 192 128 64 Decoder LSTM 1024 786 512 256 Decoder CNN 512 384 256 128

[0046] Therefore, the number of filters in the convolutional layers of the encoder and decoder and the number of hidden units in the LSTM and fully connected layers are finally compressed to 50% of the original as the network structure of the acoustic model.

[0047] The Hausa vocoder on the mobile terminal can be Griffin-Lim or the neural network vocoder Squeeze_wave. Griffin-Lim is an algorithm that reconstructs speech when only the amplitude spectrum is known but the phase spectrum is unknown. Squeeze_wave is a stream-based generative model that can map spectral features to high-quality speech. Squeeze_wave analyzes the model parameters and computational complexity of WaveGlow and adopts a more efficient computing unit. Therefore, it greatly reduces the number of model parameters and computational complexity without losing much sound quality, allowing it to achieve a very low real-time rate on mobile devices.

[0048] The Hausa speech synthesis method provided by the embodiment of the present invention performs front-end processing on Hausa text input by a user, uses conditional random fields based on the contextual semantics of the language to measure the connection between word tones, thereby improving the accuracy of subsequent synthesis. Different combinations of acoustic models and vocoders are selected for different end-to-end and mobile application scenarios, which can meet corresponding requirements, improve the accuracy, naturalness and synthesis efficiency of Hausa speech synthesis, and can be applied to various speech interaction scenarios in the Hausa context.

[0049] Example 2

[0050] The embodiment of the present invention provides a Hausa speech synthesis system, such as Figure 3 As shown, including:

[0051] The Hausa language front-end processing module 1 is used to perform polyphone disambiguation processing on the input Hausa language text. This module executes the method described in step 1 of embodiment 1, which will not be described in detail here.

[0052] The Hausa acoustic model module 2 is used to input the processed Hausa text into the Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech; this module executes the method described in step S2 of embodiment 1, which will not be repeated here.

[0053] The Hausa vocoder module 3 is used to input the spectral features into the Hausa vocoder and map the spectral features into speech. This module executes the method described in step S3 of embodiment 1, which will not be described in detail here.

[0054] The Hausa speech synthesis system provided by the embodiment of the present invention improves the accuracy of synthesis by setting a Hausa front-end processing module, improves the naturalness of synthesis by an acoustic model module, and improves the efficiency of synthesis by a vocoder module. The front-end processing module uses the contextual semantics of the language and uses a conditional random field to measure the connection between word tones. The entire recognition model has a simple structure and a small number of model parameters. It is easy to deploy on a cloud or mobile device, and can quickly and accurately calculate the recognition result for the input Hausa text and perform speech synthesis. It can be applied to various speech interaction scenarios in the Hausa language context.

[0055] Example 3

[0056] An embodiment of the present invention provides a computer device, such as Figure 4 As shown, it includes: at least one processor 401, such as a CPU (Central Processing Unit), at least one communication interface 403, a memory 404, and at least one communication bus 402. Among them, the communication bus 402 is used to realize the connection and communication between these components. Among them, the communication interface 403 may include a display screen (Display), a keyboard (Keyboard), and the optional communication interface 403 may also include a standard wired interface and a wireless interface. The memory 404 may be a high-speed RAM memory (Ramdom Access Memory, volatile random access memory) or a non-volatile memory (non-volatile memory), such as at least one disk memory. The memory 404 may also be at least one storage device located away from the aforementioned processor 401. Among them, the processor 401 can execute the method described in Example 1. A set of program codes are stored in the memory 404, and the processor 401 calls the program code stored in the memory 404 to execute the Hausa speech synthesis method of Example 1.

[0057] The communication bus 402 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The communication bus 402 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 The fact that only one line is used in the diagram does not mean that there is only one bus or only one type of bus.

[0058] The memory 404 may include a volatile memory (English: volatile memory), such as a random access memory (English: random-access memory, abbreviated: RAM); the memory may also include a non-volatile memory (English: non-volatile memory), such as a flash memory (English: flash memory), a hard disk drive (English: hard disk drive, abbreviated: HDD) or a solid-state drive (English: solid-state drive, abbreviated: SSD); the memory 404 may also include a combination of the above-mentioned types of memory. The processor 401 may be a central processing unit (English: central processing unit, abbreviated: CPU), a network processor (English: network processor, abbreviated: NP) or a combination of CPU and NP.

[0059] The processor 401 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof. Optionally, the memory 404 is also used to store program instructions. The processor 401 may call program instructions to implement the Hausa speech synthesis method of the embodiment 1 of the present application.

[0060] The embodiment of the present invention further provides a computer-readable storage medium, on which computer-executable instructions are stored, and the computer-executable instructions can execute the Hausa speech synthesis method of embodiment 1. The storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

[0061] Obviously, the above embodiments are merely examples for clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. A Hausa speech synthesis method, characterized in that: include: The input Hausa text is subjected to polyphone disambiguation processing, including: inputting the Hausa text into a deep neural network to output the tone score of each word, and obtaining the optimal tone recognition result through a decoding module, wherein the deep neural network includes: a word vector module, a context semantic extraction module, a fully connected layer, and a conditional random field module, wherein: The Hausa text to be recognized is used to obtain the pre-trained word vector of each word in the text through the word vector module, and the obtained word vector is sent to the context semantic extraction module to obtain the semantic representation of each word containing context information. After obtaining the representation of each word with context semantic information, the representation of each word is sent to the fully connected layer to obtain the score on each tone. After obtaining the transfer score between the tones through the conditional random field, combined with the previously calculated scores of each word on each tone, the decoding module is used to decode the optimal tone recognition result for the entire text; The processed Hausa text is input into a Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech. The Hausa acoustic model adopts Tacotron 2 compressed according to a preset compression ratio. The Tacotron2 includes an encoder, a decoder and an attention module, wherein the encoder includes: 1 embedding layer, 3 convolutional layers, and 1 bidirectional long short-term memory layer; the decoder includes: 2 fully connected layers, 2 unidirectional long short-term memory layers, 1 mapping layer and 5 convolutional layers. The number of filters in the convolutional layers of the encoder and decoder and the number of hidden units in the LSTM and fully connected layers are compressed to 50% of the original as the network structure of the acoustic model; The spectral features are input into a Hausa vocoder to map the spectral features into speech.

2. The Hausa speech synthesis method according to claim 1, characterized in that: The contextual semantic extraction module consists of a one-layer bidirectional long short-term memory network.

3. The Hausa speech synthesis method according to claim 1, characterized in that: The decoding module uses the Viterbi algorithm to calculate the tone sequence with the maximum score of the Hausa text.

4. The Hausa speech synthesis method according to claim 1, characterized in that: The Hausa acoustic module model adopts the sequence-to-sequence generative model Tacotron 2 with an attention mechanism, which includes three sub-modules: encoder, decoder, and post-processing network. The encoder module processes the input Hausa text into text features with contextual information, the decoder is responsible for mapping these text features into corresponding spectral features, and the post-processing network plays a role in supplementing the details in the spectral features. The Hausa vocoder adopts WaveNet or MelGAN.

5. The Hausa speech synthesis method according to claim 1, characterized in that: For the Hausa vocoder, choose either Griffin-Lim or the neural network vocoder Squeeze_wave.

6. A Hausa speech synthesis system, characterized in that: include: The Hausa language front-end processing module is used to perform polyphone disambiguation processing and digital processing on the input Hausa language text, including: inputting the Hausa language text into a deep neural network to output the tone score of each word, and obtaining the optimal tone recognition result through a decoding module. The deep neural network includes: a word vector module, a context semantic extraction module, a fully connected layer, and a conditional random field module, wherein: The Hausa text to be recognized is used to obtain the pre-trained word vector of each word in the text through the word vector module, and the obtained word vector is sent to the context semantic extraction module to obtain the semantic representation of each word containing context information. After obtaining the representation of each word with context semantic information, the representation of each word is sent to the fully connected layer to obtain the score on each tone. After obtaining the transfer score between the tones through the conditional random field, combined with the previously calculated scores of each word on each tone, the decoding module is used to decode the optimal tone recognition result for the entire text; A Hausa acoustic model module is used to input the processed Hausa text into the Hausa acoustic model to obtain the spectral features corresponding to the Hausa speech. The Hausa acoustic model adopts Tacotron 2 compressed according to a preset compression ratio. The Tacotron2 includes an encoder, a decoder and an attention module, wherein the encoder includes: 1 embedding layer, 3 convolution layers, and 1 bidirectional long short-term memory layer; the decoder includes: 2 fully connected layers, 2 unidirectional long short-term memory layers, 1 mapping layer and 5 convolution layers. The number of filters in the convolution layers of the encoder and decoder and the number of hidden units in the LSTM and fully connected layers are compressed to 50% of the original as the network structure of the acoustic model; The Hausa vocoder module is used to input the frequency spectrum features into a Hausa vocoder and map the frequency spectrum features into speech.

7. A computer device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the Hausa speech synthesis method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the Hausa speech synthesis method according to any one of claims 1 to 5.

Images