如何在Python的neuralintents库中使用intents.json里的responses

问题描述 #

使用Python的neuralintents模块开发语音助手时，发现该库仅支持将意图映射到函数，无法直接利用intents.json文件中的responses字段向用户反馈，希望找到实现该功能的方法。

解决方案 #

neuralintents官方没有直接提供调用responses的API，但可以通过手动读取intents.json并结合意图预测结果来实现，或者扩展GenericAssistant类封装逻辑。以下是具体实现方式：

方法一：手动读取intents.json并匹配回复 #

1. 加载并解析intents.json文件，将其转为Python字典
2. 使用assistant.predict_intent()获取用户输入对应的意图标签
3. 根据意图标签从字典中取出对应的responses列表，随机选择一条回复
4. 用语音合成模块播报该回复

修改后的完整代码

from neuralintents import GenericAssistant
import speech_recognition
import pyttsx3 as tts
import sys
import json
import random

# 初始化语音识别与合成
recognizer = speech_recognition.Recognizer()
speaker = tts.init()
speaker.setProperty('rate', 150)
# 可选：切换女声
voices = speaker.getProperty('voices')
speaker.setProperty('voice', voices[1].id)

todo_list = ['Go shopping', 'Clean Room', 'Record Videos']

# 加载intents.json
def load_intents(file_path):
    with open(file_path, 'r', encoding='utf-8') as f:
        return json.load(f)

intents_data = load_intents('intents.json')

# 原有功能函数保留
def create_note():
    global recognizer
    speaker.say("What do you want to write into your note")
    speaker.runAndWait()

    done = False

    while not done:
        try:
            with speech_recognition.Microphone() as mic:
                recognizer.adjust_for_ambient_noise(mic, duration=0.2)
                audio = recognizer.listen(mic)

                note = recognizer.recognize_google(audio)
                note = note.lower()

                speaker.say("Choose a filename!")
                speaker.runAndWait()

                recognizer.adjust_for_ambient_noise(mic, duration=0.2)
                audio = recognizer.listen(mic)

                filename = recognizer.recognize_google(audio)
                filename = filename.lower()

            with open(filename, 'w') as f:
                f.write(note)
                done = True
                speaker.say(f"I successfully created the note {filename}")
                speaker.runAndWait()
        except speech_recognition.UnknownValueError:
            recognizer = speech_recognition.Recognizer()
            speaker.say("I did not understand you! Please try again!")
            speaker.runAndWait()

def add_todo():
    global recognizer

    speaker.say("What todo do you want to add?")
    speaker.runAndWait()

    done = False

    while not done:
        try:
            with speech_recognition.Microphone() as mic:
                recognizer.adjust_for_ambient_noise(mic, duration=0.2)
                audio = recognizer.listen(mic)

                item = recognizer.recognize_google(audio)
                item = item.lower()

                todo_list.append(item)
                done = True

                speaker.say(f"I added {item} to the to do list!")
                speaker.runAndWait()

        except speech_recognition.UnknownValueError:
            recognizer = speech_recognition.Recognizer()
            speaker.say("I did not understand. Please try again!")
            speaker.runAndWait()

def show_todos():
    speaker.say("The items in your to do list are the following")
    for item in todo_list:
        speaker.say(item)
    speaker.runAndWait()

def exitt():
    # 从intents中获取goodbye的回复
    for intent in intents_data['intents']:
        if intent['tag'] == 'goodbye':
            response = random.choice(intent['responses'])
            speaker.say(response)
            speaker.runAndWait()
            sys.exit(0)

# 意图映射：仅保留需要执行逻辑的意图，纯回复类意图可以不用映射
mappings = {
    "create_note": create_note,
    "add_todos": add_todo,
    "show_todos": show_todos,
    "goodbye": exitt,
}

assistant = GenericAssistant('intents.json', intent_methods=mappings)
assistant.train_model()
assistant.save_model()

# 处理用户请求的核心逻辑
def handle_command(command):
    intent_tag = assistant.predict_intent(command)
    
    # 先检查是否有映射的函数，优先执行函数
    if intent_tag in mappings:
        mappings[intent_tag]()
    else:
        # 没有映射函数则从intents中取回复
        for intent in intents_data['intents']:
            if intent['tag'] == intent_tag:
                response = random.choice(intent['responses'])
                speaker.say(response)
                speaker.runAndWait()
                break

while True:
    try:
        with speech_recognition.Microphone() as mic:
            recognizer.adjust_for_ambient_noise(mic, duration=0.2)
            audio = recognizer.listen(mic)

            command = recognizer.recognize_google(audio)
            command = command.lower()
        
        handle_command(command)

    except speech_recognition.UnknownValueError:
        recognizer = speech_recognition.Recognizer()
        # 可以添加默认回复
        speaker.say("Sorry, I didn't catch that. Could you repeat?")
        speaker.runAndWait()

方法二：扩展GenericAssistant类（更优雅） #

通过继承GenericAssistant，添加获取回复的方法，封装逻辑：

from neuralintents import GenericAssistant
import json
import random

class CustomAssistant(GenericAssistant):
    def __init__(self, intents_file, intent_methods=None):
        super().__init__(intents_file, intent_methods)
        # 加载intents数据
        with open(intents_file, 'r', encoding='utf-8') as f:
            self.intents_data = json.load(f)
    
    def get_response(self, command):
        intent_tag = self.predict_intent(command)
        for intent in self.intents_data['intents']:
            if intent['tag'] == intent_tag:
                return random.choice(intent['responses'])
        return "Sorry, I don't understand that."

然后在主逻辑中使用这个自定义类：

# 替换原来的assistant初始化
assistant = CustomAssistant('intents.json', intent_methods=mappings)
assistant.train_model()
assistant.save_model()

# 在循环中调用
while True:
    try:
        with speech_recognition.Microphone() as mic:
            recognizer.adjust_for_ambient_noise(mic, duration=0.2)
            audio = recognizer.listen(mic)
            command = recognizer.recognize_google(audio).lower()
        
        # 先检查是否有映射函数
        intent_tag = assistant.predict_intent(command)
        if intent_tag in mappings:
            mappings[intent_tag]()
        else:
            response = assistant.get_response(command)
            speaker.say(response)
            speaker.runAndWait()

    except speech_recognition.UnknownValueError:
        recognizer = speech_recognition.Recognizer()
        speaker.say("Sorry, I didn't catch that. Could you repeat?")
        speaker.runAndWait()

说明 #

• 纯回复类的意图（如问候）不需要再写对应的函数，直接从intents.json中取回复即可
• 对于需要执行逻辑的意图（如创建笔记、添加待办），依然保留函数映射，执行完逻辑后也可以结合responses字段给用户反馈
• 随机选择回复可以让助手的交互更自然

内容的提问来源于stack exchange，提问作者Dinujaya Sandaruwan