Warehouse goods sorting method based on voice recognition

Through voice recognition technology, the Dixtra algorithm is used to generate the optimal picking route and convert it into voice commands, which solves the problem of high error rate of manual search routes by pickers in traditional warehouse picking, and improves the picking efficiency and accuracy.

CN119047976AInactive Publication Date: 2024-11-29四川参盘供应链科技有限公司
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411479384.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-23
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

In traditional warehouse picking operations, pickers need to manually find the picking route, resulting in an increase in walking distance and a high error rate, which affects efficiency and cost.

Method used

Using a method of warehousing goods picking based on voice recognition, the Dixtra algorithm generates the optimal picking route by obtaining order information and real-time location of pickers, and converting the route into voice commands, providing voice prompts, reducing manual operations and Error rate.

Benefits of technology

The optimization of the picking route is achieved, which reduces time and labor costs, reduces human error rates, and makes the picking process more efficient and accurate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119047976A_ABST
    Figure CN119047976A_ABST
Patent Text Reader

Abstract

The invention discloses a warehouse goods sorting method based on voice recognition, and belongs to the technical field of warehouse logistics management, and the method comprises the following steps: S1, obtaining order information and the real-time position of a sorter; s2, according to the storage position of the goods in the order information and the real-time position of the picker, generating a picking route, and determining a picking task text corresponding to the picking route; s3, converting the sorting task text into sorting voice; and S4, sending the sorting voice to a sorter, and carrying out voice prompt. According to the method, the sorting personnel are allowed to directly execute the sorting task through the voice instruction without holding the equipment, so that the time waste caused by equipment operation is reduced, and the human error rate is greatly reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of warehousing logistics management, and in particular relates to a warehousing goods picking method based on voice recognition. Background Art

[0002] In traditional warehouse picking operations, pickers need to manually find the picking route or rely on paper documents to complete the picking. Pickers may need to travel back and forth between different storage locations many times, which increases the walking distance and further reduces work efficiency. Manually finding the picking route is prone to errors, such as misreading the document and going to the wrong storage location, which will lead to picking errors and increase the complexity and cost of subsequent processing. With the development of intelligent technology, voice picking technology has gradually been applied to the field of warehousing and logistics, which can provide pickers with auxiliary instructions in the form of voice to avoid picking errors. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a warehouse goods picking method based on speech recognition.

[0004] The technical solution of the present invention is: a method for picking warehouse goods based on voice recognition comprises the following steps:

[0005] S1, obtain order information and the real-time location of the picker;

[0006] S2. Generate a picking route based on the storage location of the goods in the order information and the real-time location of the picker, and determine the picking task text corresponding to the picking route;

[0007] S3, converting the picking task text into picking voice;

[0008] S4. Send the picking voice to the picking personnel for voice prompting.

[0009] Furthermore, in S2, the Dijkstra algorithm is used to generate a picking route according to the storage location of the goods in the order information and the real-time location of the picker.

[0010] The beneficial effect of the above further solution is that in the present invention, the Dijkstra algorithm can find the shortest path from one vertex to the remaining vertices, solving the shortest path problem in the weighted graph. In S2, the Dijkstra algorithm can be used to solve the problem of picking routes, so that the routes of the pickers are optimal, reducing the picking time cost and labor cost. The picking routes are then translated into text form for easy conversion into speech.

[0011] Furthermore, S3 includes the following sub-steps:

[0012] S31, performing word segmentation processing on the picking task text, and performing part-of-speech tagging on each word obtained by the word segmentation processing;

[0013] S32, inputting the part of speech of each word into the phonological feature function to obtain the phonological feature coefficient of each word;

[0014] S33, inputting each word into the speech conversion model, and generating a mel-spectrogram using the phonological feature coefficient of each word;

[0015] S34. Convert the picking task text into picking speech using the Mel-spectrogram.

[0016] The beneficial effect of the above further scheme is: in the present invention, after the parts of speech are marked on the words, the parts of speech such as prepositions, adverbs, nouns and conjunctions can be obtained. The pronunciations of different parts of speech are different. Therefore, the present invention needs to generate a phonological feature coefficient for the word according to the part-of-speech characteristics of the word. The phonological feature coefficient can affect the tone of the word in the entire selected speech to ensure that the generated speech is more fluent.

[0017] Furthermore, in S32, the expression of the phonological feature function Y is:

[0018] ; In the formula, η represents the weight of the real or virtual word, f represents the frequency of the word in the picking task text, and c represents a constant.

[0019] In the present invention, if the word is a content word (such as nouns and verbs), it has actual meaning and can serve as a sentence component alone, so its phonological characteristic coefficient should be large, η = 2. On the contrary, if the word is a function word (such as adverbs and prepositions), it has no complete meaning, and its phonological characteristic coefficient should be small, η = 1.

[0020] Further, in S33, the speech conversion model includes an input unit, a classifier, a plurality of recurrent units, an operator and an output unit;

[0021] The input end of the input unit is used as the input end of the speech conversion model; the output end of the input unit is connected to the input end of the classifier; the output end of the classifier is connected to the input end of several cyclic units; the output ends of several cyclic units are connected to the input end of the operator; the output end of the operator is connected to the input end of the output unit; the output end of the output unit is used as the output end of the speech conversion model.

[0022] The beneficial effect of the above further scheme is: in the present invention, the input unit is used to input each word into the speech conversion model, the classifier segments the picking task text to obtain several complete sentences, and each complete sentence is input into the loop unit for data processing. The operator integrates the phonological feature coefficient of each word into the output result of each loop unit to ensure that the Mel spectrum output by the output unit is accurate.

[0023] Furthermore, the expression of the output C of the recurrent unit is:

[0024] ; In the formula, I represents the number of words contained in the sentence processed by the recurrent unit, Q i represents the word embedding vector of the i-th word in the sentence processed by the recurrent unit, W represents the weight matrix of the recurrent unit, b represents the bias matrix of the recurrent unit, Relu(·) represents the activation function, exp(·) represents the exponential function, and f i Represents the frequency of the i-th word in the sentence processed by the recurrent unit in the picking task text.

[0025] The beneficial effect of the above further solution is that in the present invention, there are weight terms and bias terms in the cyclic unit, both of which are part of the neural network parameters and are used to increase the flexibility of the network.

[0026] Furthermore, the expression of the output M of the operator is:

[0027] Where, C j represents the output of the jth recurrent unit, Y j,1 represents the phonological feature coefficient of the first word in the sentence processed by the jth recurrent unit, Y j,i represents the phonological feature coefficient of the i-th word in the sentence processed by the j-th recurrent unit, Y j,I represents the phonological feature coefficient of the Ith word in the sentence processed by the jth recurrent unit, max(·) represents the maximum value function, represents the upward rounding function, J represents the number of recurrent units, and α represents the threshold value of the neuron node corresponding to the operator.

[0028] The beneficial effect of the above further scheme is: in the present invention, in the speech conversion model, the operator is composed of neuron nodes, and the threshold is a boundary value used to determine whether the neuron should be activated, and should be used as an important parameter to participate in the operator's calculation of the output result of the loop layer.

[0029] Furthermore, in S32, a vocoder is used to convert the picking task text into picking speech based on the mel-spectrogram.

[0030] The beneficial effect of the above further solution is that: in the present invention, a vocoder is used to convert the Mel-spectrogram into a time-domain speech waveform, and the converted time-domain speech waveform is output as an audio file for real-time playback.

[0031] The beneficial effects of the present invention are as follows: the present invention matches the picking task text corresponding to the order information with the corresponding picking voice, and the picking voice fully considers the characteristics of the words themselves, so that the generated picking voice is fluent and natural, and converts the picking route into voice broadcast to the pickers, and guides the pickers to pick up the goods at the designated storage location through voice. This method allows the pickers to perform picking tasks directly through voice commands without the need for handheld devices, thereby reducing the time waste caused by equipment operation and greatly reducing the human error rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of a method for picking warehouse goods based on speech recognition;

[0033] Figure 2 Schematic diagram of the structure of the speech conversion model. DETAILED DESCRIPTION

[0034] The embodiments of the present invention will be further described below in conjunction with the accompanying drawings.

[0035] like Figure 1 As shown, the present invention provides a method for picking warehouse goods based on voice recognition, comprising the following steps:

[0036] S1, obtain order information and the real-time location of the picker;

[0037] S2. Generate a picking route based on the storage location of the goods in the order information and the real-time location of the picker, and determine the picking task text corresponding to the picking route;

[0038] S3, converting the picking task text into picking voice;

[0039] S4. Send the picking voice to the picking personnel for voice prompting.

[0040] In the embodiment of the present invention, in S2, the picking route is generated using the Dijkstra algorithm according to the storage location of the goods in the order information and the real-time location of the picker.

[0041] In the present invention, the Dijkstra algorithm can find the shortest path from one vertex to the remaining vertices, solving the shortest path problem in a weighted graph. In S2, the Dijkstra algorithm can be used to solve the problem of picking routes, so that the routes of pickers are optimal, reducing the time cost and labor cost of picking. The picking routes are then translated into text form for easy conversion into speech.

[0042] In this embodiment of the present invention, S3 includes the following sub-steps:

[0043] S31, performing word segmentation processing on the picking task text, and performing part-of-speech tagging on each word obtained by the word segmentation processing;

[0044] S32, inputting the part of speech of each word into the phonological feature function to obtain the phonological feature coefficient of each word;

[0045] S33, inputting each word into the speech conversion model, and generating a mel-spectrogram using the phonological feature coefficient of each word;

[0046] S34. Convert the picking task text into picking speech using the Mel-spectrogram.

[0047] In the present invention, after the parts of speech of words are marked, the parts of speech such as prepositions, adverbs, nouns and conjunctions can be obtained. The pronunciations of different parts of speech are different. Therefore, the present invention needs to generate a phonological feature coefficient for the word according to the part-of-speech characteristics of the word. The phonological feature coefficient can affect the tone of the word in the entire selected speech to ensure that the generated speech is more fluent.

[0048] In the embodiment of the present invention, in S32, the expression of the phonological feature function Y is: ; In the formula, η represents the weight of the real or virtual word, f represents the frequency of the word in the picking task text, and c represents a constant.

[0049] In the present invention, if the word is a content word (such as nouns and verbs), it has actual meaning and can serve as a sentence component alone, so its phonological characteristic coefficient should be large, η = 2. On the contrary, if the word is a function word (such as adverbs and prepositions), it has no complete meaning, and its phonological characteristic coefficient should be small, η = 1.

[0050] In the embodiment of the present invention, in S33, if Figure 2 As shown, the speech conversion model includes an input unit, a classifier, a plurality of recurrent units, an operator and an output unit;

[0051] The input end of the input unit is used as the input end of the speech conversion model; the output end of the input unit is connected to the input end of the classifier; the output end of the classifier is connected to the input end of several cyclic units; the output ends of several cyclic units are connected to the input end of the operator; the output end of the operator is connected to the input end of the output unit; the output end of the output unit is used as the output end of the speech conversion model.

[0052] In the present invention, the input unit is used to input each word into the speech conversion model, the classifier segments the picking task text to obtain several complete sentences, and each complete sentence is input into the loop unit for data processing. The operator integrates the phonological feature coefficients of each word into the output results of each loop unit to ensure that the Mel-spectrogram output by the output unit is accurate.

[0053] In the embodiment of the present invention, the expression of the output C of the circulation unit is:

[0054] ; In the formula, I represents the number of words contained in the sentence processed by the recurrent unit, Q i represents the word embedding vector of the i-th word in the sentence processed by the recurrent unit, W represents the weight matrix of the recurrent unit, b represents the bias matrix of the recurrent unit, Relu(·) represents the activation function, exp(·) represents the exponential function, and f i Represents the frequency of the i-th word in the sentence processed by the recurrent unit in the picking task text.

[0055] In the present invention, there are weight terms and bias terms in the cyclic unit, both of which are part of the neural network parameters and are used to increase the flexibility of the network.

[0056] In the embodiment of the present invention, the expression of the output M of the operator is:

[0057] Where, C j represents the output of the jth recurrent unit, Y j,1 represents the phonological feature coefficient of the first word in the sentence processed by the jth recurrent unit, Y j,i represents the phonological feature coefficient of the i-th word in the sentence processed by the j-th recurrent unit, Y j,I represents the phonological feature coefficient of the Ith word in the sentence processed by the jth recurrent unit, max(·) represents the maximum value function, represents the upward rounding function, J represents the number of recurrent units, and α represents the threshold value of the neuron node corresponding to the operator.

[0058] In the present invention, in the speech conversion model, the operator is composed of neuron nodes, and the threshold is a boundary value used to determine whether the neuron should be activated. It should be used as an important parameter to participate in the operator's calculation of the output result of the loop layer.

[0059] In the embodiment of the present invention, in S32, a vocoder is used to convert the picking task text into picking speech based on the mel-spectrogram.

[0060] In the present invention, a vocoder is used to convert the mel-spectrogram into a time-domain speech waveform, and the converted time-domain speech waveform is output as an audio file for real-time playback.

[0061] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A method for picking warehouse goods based on speech recognition, characterized in that: The following steps are involved: S1, obtain order information and the real-time location of the picker; S2. Generate a picking route based on the storage location of the goods in the order information and the real-time location of the picker, and determine the picking task text corresponding to the picking route; S3, converting the picking task text into picking voice; S4. Send the picking voice to the picking personnel for voice prompting.

2. The method for picking warehouse goods based on voice recognition according to claim 1, characterized in that: In S2, the picking route is generated using the Dijkstra algorithm according to the storage location of the goods in the order information and the real-time location of the picker.

3. The method for picking warehouse goods based on voice recognition according to claim 1, characterized in that: The S3 comprises the following sub-steps: S31, performing word segmentation processing on the picking task text, and performing part-of-speech tagging on each word obtained by the word segmentation processing; S32, inputting the part of speech of each word into the phonological feature function to obtain the phonological feature coefficient of each word; S33, inputting each word into the speech conversion model, and generating a mel-spectrogram using the phonological feature coefficient of each word; S34. Convert the picking task text into picking speech using the Mel-spectrogram.

4. The method for picking warehouse goods based on voice recognition according to claim 3, characterized in that: In S32, the expression of the phonological feature function Y is: ; In the formula, η represents the weight of the real or virtual word, f represents the frequency of the word in the picking task text, and c represents a constant.

5. The method for picking warehouse goods based on voice recognition according to claim 3, characterized in that: In S33, the speech conversion model includes an input unit, a classifier, a plurality of recurrent units, an operator and an output unit; The input end of the input unit serves as the input end of the speech conversion model; the output end of the input unit is connected to the input end of the classifier; the output end of the classifier is connected to the input end of several cycle units; the output ends of several cycle units are connected to the input end of the operator; the output end of the operator is connected to the input end of the output unit; the output end of the output unit serves as the output end of the speech conversion model.

6. The method for picking warehouse goods based on voice recognition according to claim 5, characterized in that: The expression of the output C of the recurrent unit is: ; In the formula, I represents the number of words contained in the sentence processed by the recurrent unit, Q i represents the word embedding vector of the i-th word in the sentence processed by the recurrent unit, W represents the weight matrix of the recurrent unit, b represents the bias matrix of the recurrent unit, Relu(·) represents the activation function, exp(·) represents the exponential function, and f i Represents the frequency of the i-th word in the sentence processed by the recurrent unit in the picking task text.

7. The method for picking warehouse goods based on voice recognition according to claim 5, characterized in that: The expression of the output M of the operator is: Where, C j represents the output of the jth recurrent unit, Y j,1 represents the phonological feature coefficient of the first word in the sentence processed by the jth recurrent unit, Y j,i represents the phonological feature coefficient of the i-th word in the sentence processed by the j-th recurrent unit, Y j,I represents the phonological feature coefficient of the Ith word in the sentence processed by the jth recurrent unit, max(·) represents the maximum value function, represents the upward rounding function, J represents the number of recurrent units, and α represents the threshold value of the neuron node corresponding to the operator.

8. The method for picking warehouse goods based on voice recognition according to claim 3, characterized in that: In S32, a vocoder is used to convert the picking task text into picking speech based on the mel-spectrogram.

Citation Information

Patent Citations

  • Picking system and picking method

    CN110874669A

  • Virtual teacher simulation voice generation method and terminal

    CN115662435A

  • Voice processing method and device for intelligent customer service, equipment and medium

    CN118298804A

  • Mobile-assisted picker techniques for in-store navigation

    US20230005046A1

  • Method and system for speech broadcast of text

    WO2018121757A1