Improve TTS

- Add pausing (fixes #727)
- Add "speak from here"
- Fix SSIP client
- Autospawn speech-dispatcher

src/book-viewer.js

@@ -12,7 +12,7 @@ import { gettext as _ } from 'gettext'
 import * as utils from './utils.js'
 import * as format from './format.js'
 import { WebView } from './webview.js'
-import * as speech from './speech.js'
+import { SSIPClient } from './speech.js'
 
 import './toc.js'
 import './search.js'
@@ -25,6 +25,8 @@ import { ImageViewer } from './image-viewer.js'
 import { makeBookInfoWindow } from './book-info.js'
 import { getURIStore, getBookList } from './library.js'
 
+const ssip = new SSIPClient()
+
 // for use in the WebView
 const uiText = {
     close: _('Close'),
@@ -573,7 +575,9 @@ GObject.registerClass({
     addAnnotation(x) { return this.#exec('reader.view.addAnnotation', x) }
     deleteAnnotation(x) { return this.#exec('reader.view.deleteAnnotation', x) }
     print() { return this.#exec('reader.print') }
+    initSpeech(x) { return this.#exec('reader.view.initSpeech', x) }
     speak(x) { return this.#exec('reader.view.speak', x) }
+    resumeSpeech() { return this.#exec('reader.view.resumeSpeech') }
     hightlightSpeechMark(x) { return this.#exec('reader.view.hightlightSpeechMark', x) }
     getCover() { return this.#exec('reader.getCover').then(utils.base64ToPixbuf) }
     init(x) { return this.#exec('reader.view.init', x) }
@@ -634,6 +638,7 @@ export const BookViewer = GObject.registerClass({
     #book
     #cover
     #data
+    #ttsPaused
     constructor(params) {
         super(params)
         utils.connect(this._view, {
@@ -812,7 +817,7 @@ export const BookViewer = GObject.registerClass({
                 'toggle-toc', 'toggle-annotations', 'toggle-bookmarks',
                 'preferences', 'show-info', 'bookmark',
                 'export-annotations',
-                'tts-speak', 'tts-stop',
+                'tts-speak', 'tts-pause', 'tts-stop',
             ],
             props: ['fold-sidebar'],
         })
@@ -991,6 +996,7 @@ export const BookViewer = GObject.registerClass({
                     this._search_view.doSearch()
                 },
                 'print': () => resolve('print'),
+                'speak-from-here': () => resolve('speak-from-here'),
             }))
             utils.connect(popover, {
                 'show-popover': (_, popover) =>
@@ -1030,6 +1036,8 @@ export const BookViewer = GObject.registerClass({
                         : format.vprintf(_('‘%s’'), [text])
             utils.setClipboardText(result, this.root)
         }
+        else if (action === 'speak-from-here')
+            this.#speak(payload.mark).catch(e => console.error(e))
     }
     #createOverlay({ index }) {
         if (!this.#data) return
@@ -1133,18 +1141,36 @@ export const BookViewer = GObject.registerClass({
     exportAnnotations() {
         exportAnnotations(this.get_root(), this.#data.storage.export())
     }
-    async #speak() {
-        const ssml = await this._view.speak('word')
-        const iter = await speech.speak(ssml)
-        for await (const { mark } of iter) {
+    async #speak(mark) {
+        await this._view.initSpeech('word')
+        const ssml = await (!mark && this.#ttsPaused
+            ? this._view.resumeSpeech()
+            : this._view.speak(mark))
+
+        const iter = await ssip.speak(ssml)
+        this.#ttsPaused = false
+
+        let state
+        for await (const { mark, message } of iter) {
             if (mark) await this._view.hightlightSpeechMark(mark)
+            else state = message
+        }
+        if (state === 'END') {
+            // FIXME: check if at end
+            await this._view.next()
+            return this.#speak()
         }
     }
     ttsSpeak() {
         this.#speak().catch(e => console.error(e))
     }
+    ttsPause() {
+        ssip.stop()
+            .then(() => this.#ttsPaused = true)
+            .catch(e => console.error(e))
+    }
     ttsStop() {
-        speech.stop().catch(e => console.error(e))
+        ssip.stop().catch(e => console.error(e))
     }
     vfunc_unroot() {
         this._view.viewSettings.unbindSettings()

src/reader/reader.js

@@ -444,6 +444,12 @@ class Reader {
                 case 'print':
                     this.printRange(range.startContainer.ownerDocument, range)
                     break
+                case 'speak-from-here':
+                    this.view.initSpeech('word').then(() => emit({
+                        type: 'selection', action,
+                        mark: this.view.getSpeechMarkBefore(range),
+                    }))
+                    break
             }
         })
     }

src/speech.js

@@ -1,19 +1,28 @@
 import Gio from 'gi://Gio'
 import GLib from 'gi://GLib'
 
-class SSIPClient {
+class SSIPConnection {
+    #connection
     #inputStream
     #outputStream
     #onResponse
     #eventData = []
     constructor(onEvent) {
         this.onEvent = onEvent
+    }
+    spawn() {
+        const flags = Gio.SubprocessFlags.NONE
+        const launcher = new Gio.SubprocessLauncher({ flags })
+        const proc = launcher.spawnv(['speech-dispatcher', '--spawn'])
+        return new Promise(resolve => proc.wait_check_async(null, () => resolve()))
+    }
+    connect() {
         const path = GLib.build_filenamev(
             [GLib.get_user_runtime_dir(), 'speech-dispatcher/speechd.sock'])
         const address = Gio.UnixSocketAddress.new(path)
-        const connection = new Gio.SocketClient().connect(address, null)
-        this.#outputStream = Gio.DataOutputStream.new(connection.get_output_stream())
-        this.#inputStream = Gio.DataInputStream.new(connection.get_input_stream())
+        this.#connection = new Gio.SocketClient().connect(address, null)
+        this.#outputStream = Gio.DataOutputStream.new(this.#connection.get_output_stream())
+        this.#inputStream = Gio.DataInputStream.new(this.#connection.get_input_stream())
         this.#inputStream.newline_type = Gio.DataStreamNewlineType.TYPE_CR_LF
         this.#receive()
     }
@@ -38,6 +47,7 @@ class SSIPClient {
     }
     send(command) {
         return new Promise((resolve, reject) => {
+            if (!this.#connection.is_connected()) reject()
             this.#outputStream.put_string(command + '\r\n', null)
             const data = []
             this.#onResponse = (code, end, message) => {
@@ -50,16 +60,25 @@ class SSIPClient {
     }
 }
 
-class SpeechD {
+export class SSIPClient {
+    #initialized
     #promises = new Map()
-    #client = new SSIPClient((msgID, result) =>
+    #connection = new SSIPConnection((msgID, result) =>
         this.#promises.get(msgID)?.resolve?.(result))
     async init() {
-        const clientName = `${GLib.get_user_name()}:foliate:tts`
-        await this.#client.send('SET SELF CLIENT_NAME ' + clientName)
-        await this.#client.send('SET SELF SSML_MODE on')
-        await this.#client.send('SET SELF NOTIFICATION ALL on')
-        return this
+        if (this.#initialized) return
+        this.#initialized = true
+        try {
+            await this.#connection.spawn()
+            this.#connection.connect()
+            const clientName = `${GLib.get_user_name()}:foliate:tts`
+            await this.#connection.send('SET SELF CLIENT_NAME ' + clientName)
+            await this.#connection.send('SET SELF SSML_MODE on')
+            await this.#connection.send('SET SELF NOTIFICATION ALL on')
+        } catch (e) {
+            this.#initialized = false
+            throw e
+        }
     }
     #makePromise(msgID){
         return new Promise((resolve, reject) => this.#promises.set(msgID, {
@@ -81,10 +100,14 @@ class SpeechD {
             },
         }
     }
+    async send(command) {
+        await this.init()
+        return this.#connection.send(command)
+    }
     async speak(str) {
-        await this.#client.send('SPEAK')
+        await this.send('SPEAK')
         const text = str.replace('\r\n.', '\r\n..') + '\r\n.'
-        const [msgID] = await this.#client.send(text)
+        const [msgID] = await this.send(text)
         const iter = this.#makeIter(msgID)
         let done = false
         const next = async () => {
@@ -103,10 +126,20 @@ class SpeechD {
             [Symbol.asyncIterator]: () => ({ next }),
         }
     }
+    pause() {
+        return this.send('PAUSE all')
+    }
+    resume() {
+        return this.send('RESUME all')
+    }
     stop() {
-        return this.#client.send('STOP all')
+        return this.send('STOP all')
+    }
+    async listSynthesisVoices() {
+        const data = await this.send('LIST SYNTHESIS_VOICES')
+        return data.map(row => {
+            const [name, lang, variant] = row.split('\t')
+            return { name, lang, variant }
+        })
     }
 }
-
-export const speak = str => new SpeechD().init().then(x => x.speak(str))
-export const stop = () => new SpeechD().init().then(x => x.stop())

src/ui/navbar.ui

@@ -13,12 +13,17 @@
   <section>
     <attribute name="display-hint">horizontal-buttons</attribute>
     <item>
-      <attribute name="label" translatable="yes">Start Speaking</attribute>
+      <attribute name="label" translatable="yes">Start</attribute>
       <attribute name="action">viewer.tts-speak</attribute>
       <attribute name="verb-icon">media-playback-start-symbolic</attribute>
     </item>
     <item>
-      <attribute name="label" translatable="yes">Stop Speaking</attribute>
+      <attribute name="label" translatable="yes">Pause</attribute>
+      <attribute name="action">viewer.tts-pause</attribute>
+      <attribute name="verb-icon">media-playback-pause-symbolic</attribute>
+    </item>
+    <item>
+      <attribute name="label" translatable="yes">Stop</attribute>
       <attribute name="action">viewer.tts-stop</attribute>
       <attribute name="verb-icon">media-playback-stop-symbolic</attribute>
     </item>