Enhance audio processing capabilities in LLM routes

- Added support for transcription and speech generation using multiple providers (OpenAI, Google, ElevenLabs).
- Introduced default configurations for transcription and speech services.
- Updated the aiService to handle different providers for audio transcription and speech synthesis.
- Implemented file storage for generated audio files with expiration handling.
- Improved error handling for integration key retrieval and provider validation.

Author: JohnnyUB

routes/llm.js

@@ -6,6 +6,12 @@ const aiService = require('../services/aiService');
 const multer = require('multer');
 const fileUtils = require('../utils/fileUtils');
 const { MODELS_MULTIPLIER } = require('../utils/aiUtils');
+const mongoose = require('mongoose');
+const uuidv4 = require('uuid/v4');
+const FileGridFsService = require('../services/fileGridFsService');
+
+const fileService = new FileGridFsService('files');
+const chatFileExpirationTime = parseInt(process.env.CHAT_FILE_EXPIRATION_TIME || '2592000', 10);
 
 let MAX_UPLOAD_FILE_SIZE = process.env.MAX_UPLOAD_FILE_SIZE;
 let uploadlimits = undefined;
@@ -18,6 +24,20 @@ if (MAX_UPLOAD_FILE_SIZE) {
 }
 var upload = multer({limits: uploadlimits});
 
+const TRANSCRIPTION_DEFAULTS = {
+    provider: 'openai',
+    model: 'whisper-1',
+    voice: 'alloy',
+    language: 'en'
+};
+
+const SPEECH_DEFAULTS = {
+    provider: 'openai',
+    model: 'tts-1',
+    voice: 'coral',
+    language: 'en'
+};
+
 router.post('/preview', async (req, res) => {
 
     let id_project = req.projectid;
@@ -122,32 +142,68 @@ router.post('/transcription', upload.single('uploadFile'), async (req, res) => {
 
     let id_project = req.projectid;
 
+    const provider = (req.body.provider || TRANSCRIPTION_DEFAULTS.provider).toLowerCase();
+    const model = req.body.model || TRANSCRIPTION_DEFAULTS.model;
+    const voice = req.body.voice || TRANSCRIPTION_DEFAULTS.voice;
+    const language = req.body.language !== undefined && req.body.language !== null
+        ? req.body.language
+        : TRANSCRIPTION_DEFAULTS.language;
+
     let file;
+    let contentType = 'audio/mpeg';
+    let filename = 'audiofile';
     if (req.body.url) {
         file = await fileUtils.downloadFromUrl(req.body.url);
     } else if (req.file) {
         file = req.file.buffer;
+        contentType = req.file.mimetype || contentType;
+        filename = req.file.originalname || filename;
     } else {
         return res.status(400).send({ success: false, error: "No audio file or URL provided"})
     }
 
     let key;
 
-    let integration = await Integration.findOne({ id_project: id_project, name: 'openai' }).catch((err) => {
-        winston.error("Error finding integration for openai");
-        return res.status(500).send({ success: false, error: "Error finding integration for openai"});
-    })
-    if (!integration) {
-        winston.verbose("Integration for openai not found.")
-        return res.status(404).send({ success: false, error: "Integration for openai not found."})
+    let integration;
+    try {
+        integration = await Integration.findOne({ id_project: id_project, name: provider });
+    } catch (err) {
+        winston.error("Error finding integration for " + provider);
+        return res.status(500).send({ success: false, error: "Error finding integration for " + provider});
     }
-    if (!integration?.value?.apikey) {
-        return res.status(422).send({ success: false, error: "The key provided for openai is not valid or undefined." })
+    if (!integration) {
+        winston.verbose("Integration for " + provider + " not found.")
+        if (provider === 'openai') {
+            winston.verbose("Try to retrieve shared OpenAI key for transcription")
+            if (!process.env.GPTKEY) {
+                winston.error("Shared key for OpenAI not configured.");
+                return res.status(404).send({ success: false, error: "No key found for " + provider });
+            }
+            key = process.env.GPTKEY;
+            winston.verbose("Using shared OpenAI key as fallback for transcription.");
+        } else {
+            return res.status(404).send({ success: false, error: "Integration for " + provider + " not found." })
+        }
+    } else if (!integration?.value?.apikey) {
+        if (provider === 'openai' && process.env.GPTKEY) {
+            key = process.env.GPTKEY;
+            winston.verbose("Using shared OpenAI key (integration key missing) for transcription.");
+        } else {
+            return res.status(422).send({ success: false, error: "The key provided for " + provider + " is not valid or undefined." })
+        }
+    } else {
+        key = integration.value.apikey;
     }
 
-    key = integration.value.apikey;
-
-    aiService.transcription(file, key).then((response) => {
+    aiService.transcription(file, {
+        key,
+        provider,
+        model,
+        voice,
+        language,
+        filename,
+        contentType
+    }).then((response) => {
         winston.verbose("Transcript response: ", response.data);
         res.status(200).send({ text: response.data.text});
     }).catch((err) => {
@@ -157,5 +213,95 @@ router.post('/transcription', upload.single('uploadFile'), async (req, res) => {
 
 })
 
+router.post('/speech', async (req, res) => {
+
+    let id_project = req.projectid;
+
+    const provider = (req.body.provider || SPEECH_DEFAULTS.provider).toLowerCase();
+    const model = req.body.model || SPEECH_DEFAULTS.model;
+    const voice = req.body.voice || SPEECH_DEFAULTS.voice;
+    const language = req.body.language !== undefined && req.body.language !== null
+        ? req.body.language
+        : SPEECH_DEFAULTS.language;
+
+    let text = req.body.text;
+
+    if (!text) {
+        return res.status(400).send({ success: false, error: "No text provided"})
+    }
+    
+    let key;
+
+    let integration;
+    try {
+        integration = await Integration.findOne({ id_project: id_project, name: provider });
+    } catch (err) {
+        winston.error("Error finding integration for " + provider);
+        return res.status(500).send({ success: false, error: "Error finding integration for " + provider});
+    }
+    if (!integration) {
+        winston.verbose("Integration for " + provider + " not found.")
+        if (provider === 'openai') {
+            winston.verbose("Try to retrieve shared OpenAI key for speech")
+            if (!process.env.GPTKEY) {
+                winston.error("Shared key for OpenAI not configured.");
+                return res.status(404).send({ success: false, error: "No key found for " + provider });
+            }
+            key = process.env.GPTKEY;
+    
+        }
+    } else if (!integration?.value?.apikey) {
+        if (provider === 'openai' && process.env.GPTKEY) {
+            key = process.env.GPTKEY;
+            winston.verbose("Using shared OpenAI key (integration key missing) for speech.");
+        } else {
+            return res.status(422).send({ success: false, error: "The key provided for " + provider + " is not valid or undefined." })
+        }
+    } else {
+        key = integration.value.apikey;
+    }
+
+    try {
+        const response = await aiService.speech(text, {
+            key,
+            provider,
+            model,
+            voice,
+            language,
+            response_format: req.body.response_format
+        });
+        const audioBuffer = response.data;
+        const contentType = response.contentType || 'audio/mpeg';
+        const ext = (response.extension || 'mp3').replace(/^\./, '');
+
+        const expireAt = new Date(Date.now() + chatFileExpirationTime * 1000);
+        var subfolder = '/public';
+        if (req.user && req.user.id) {
+            subfolder = '/users/' + req.user.id;
+        }
+        const folder = uuidv4();
+        const filePath = `uploads${subfolder}/files/${folder}/speech.${ext}`;
+
+        await fileService.createFile(filePath, audioBuffer, undefined, contentType, {
+            metadata: { expireAt }
+        });
+        const fileRecord = await fileService.find(filePath);
+        await mongoose.connection.db.collection('files.chunks').updateMany(
+            { files_id: fileRecord._id },
+            { $set: { 'metadata.expireAt': expireAt } }
+        );
+
+        winston.verbose('Speech audio stored at:', filePath);
+        return res.status(201).send({
+            message: 'Speech audio saved successfully',
+            filename: encodeURIComponent(filePath),
+            contentType
+        });
+    } catch (err) {
+        winston.error('Speech error: ', err.response?.data || err);
+        return res.status(500).send({ success: false, error: err.response?.data || err.message || err });
+    }
+})
+
 
 module.exports = router;