Text to Speech page: Speak Text section with voice menu and markup guide

AntennaHead/Services/AntennaHeadHTTPServer.swift

@@ -765,6 +765,13 @@ final class AntennaHeadHTTPServer {
                                             sequence: sequence, repeatForever: repeatForever)
             return okResponse()
 
+        case "/speaktextbuttonclicked.html":
+            // Body: {text, voice}. The Text to Speech page's Speak Text
+            // section; `voice` "" = the Text to Speech voice setting.
+            let so = jsonObject(fromBody: request.body)
+            sdrController?.speakText(so.string("text"), voiceIdentifier: so.string("voice"))
+            return okResponse()
+
         case "/playaudiofileslistenbuttonclicked.html":
             // Same payload shape as Text to Speech, plus an optional `playlist`
             // name: {sequence, repeat, files, playlist}. A non-empty playlist
@@ -2459,9 +2466,69 @@ final class AntennaHeadHTTPServer {
         s += "onclick=\"textToSpeechListenButtonClicked(getElementById('textToSpeechForm'));\" "
         s += "title='Synthesize the selected folder's text files and stream them through the live audio pipeline.'>"
         s += "</form><br>&nbsp;<br>"
+        s += speakTextFormHTML()
+        return s
+    }
+
+    /// The SSML example the Speak Text box starts with.
+    nonisolated static let speakTextExample =
+        #"<speak>It is four oh nine <break time="700ms"/> <prosody rate="80%">on AntennaHead Radio.</prosody></speak>"#
+
+    /// "Speak Text" section at the bottom of the Text to Speech page: type
+    /// text, pick a voice, Speak → `/speaktextbuttonclicked.html` →
+    /// `SDRController.speakText`. Text starting with `<speak` is SSML. The
+    /// markup guide below reflects what was measured with the Premium Ava
+    /// voice on macOS 27 (rendered clip length/content), not just Apple's docs.
+    @MainActor private func speakTextFormHTML() -> String {
+        var s = "<form class='speak_text_form' id='speakTextForm' onsubmit='event.preventDefault(); return false;' method='POST'>"
+        s += "<label for='tts_speak_text'>Speak Text</label>"
+        s += "<p>Type text and speak it once through the live audio pipeline. Text that starts with "
+        s += "<code>&lt;speak&gt;</code> is read as SSML markup (see below).</p>"
+        s += "<textarea id='tts_speak_text' name='tts_speak_text' class='u-full-width' rows='6' "
+        s += "\(Self.verbatimInputAttributes)>\(htmlText(Self.speakTextExample))</textarea>"
+        s += "<label for='tts_speak_voice'>Voice</label>"
+        s += installedVoicesSelectOptionsHTML(selectID: "tts_speak_voice")
+        s += "<br><br><input class='twelve columns button button-primary' type='button' value='Speak' "
+        s += "onclick=\"speakTextButtonClicked(getElementById('speakTextForm'));\" "
+        s += "title='Synthesize this text with the chosen voice and stream it through the live audio pipeline.'>"
+        s += "</form><br>"
+        s += Self.speechMarkupHelpHTML
+        s += "<br>&nbsp;<br>"
         return s
     }
 
+    nonisolated static let speechMarkupHelpHTML = """
+        <details class='speech-markup-help'><summary>Controlling the voice: pauses, speed, pronunciation</summary>
+        <p>There are two ways to control how the text is spoken, and they depend on the voice.</p>
+        <p><strong>Modern voices</strong> (Ava, Samantha, Siri and other Premium/Enhanced voices): start the text with
+        <code>&lt;speak&gt;</code> and end it with <code>&lt;/speak&gt;</code> to use SSML markup. Measured with the Premium Ava voice:</p>
+        <table class='u-full-width'>
+        <thead><tr><th>Markup</th><th>Effect</th></tr></thead>
+        <tbody>
+        <tr><td><code>&lt;prosody rate="50%"&gt;&hellip;&lt;/prosody&gt;</code> (also <code>"150%"</code>, <code>"slow"</code>, <code>"fast"</code>)</td><td>✅ Speed. 50% took 3.7&nbsp;s where normal took 2.8&nbsp;s.</td></tr>
+        <tr><td><code>&lt;break time="1500ms"/&gt;</code></td><td>✅ A pause of that length.</td></tr>
+        <tr><td><code>&lt;prosody volume="x-soft"&gt;&hellip;&lt;/prosody&gt;</code></td><td>✅ Volume. <code>x-soft</code> is about a quarter as loud.</td></tr>
+        <tr><td><code>&lt;say-as interpret-as="characters"&gt;KARK&lt;/say-as&gt;</code></td><td>✅ Spells it out letter by letter.</td></tr>
+        <tr><td><code>&lt;phoneme alphabet="ipa" ph="&hellip;"&gt;word&lt;/phoneme&gt;</code></td><td>✅ Fixes a pronunciation, written in the IPA phonetic alphabet.</td></tr>
+        <tr><td><code>&lt;prosody pitch="+40%"&gt;&hellip;&lt;/prosody&gt;</code></td><td>⚠️ Changes the audio only slightly; the pitch barely moves.</td></tr>
+        <tr><td><code>&lt;emphasis level="strong"&gt;&hellip;&lt;/emphasis&gt;</code></td><td>❌ Ignored.</td></tr>
+        <tr><td><code>&lt;sub alias="North Little Rock"&gt;NLR&lt;/sub&gt;</code></td><td>❌ Ignored: still says the letters. Type the words out instead.</td></tr>
+        </tbody></table>
+        <p>If the markup can&rsquo;t be parsed (for example, a closing tag that doesn&rsquo;t match its opening tag),
+        nothing is spoken at all and the log shows &ldquo;invalid SSML&rdquo;. A <code>&lt;voice name="&hellip;"&gt;</code> tag inside the markup overrides the Voice menu.</p>
+        <p><strong>Classic voices</strong> (names in the Voice menu such as Alex, Albert and Fred, whose identifiers start with
+        <code>com.apple.speech.synthesis.voice.</code>) don&rsquo;t read SSML. Instead, put commands in double brackets
+        in plain text:</p>
+        <ul>
+        <li><code>[[slnc 1500]]</code> &mdash; pause 1.5 seconds</li>
+        <li><code>[[rate 120]]</code> &mdash; speed in words per minute</li>
+        <li><code>[[pbas 40]]</code> &mdash; base pitch; <code>[[pmod 60]]</code> &mdash; how much the pitch varies</li>
+        </ul>
+        <p>Don&rsquo;t mix the two: modern voices read <code>[[&hellip;]]</code> commands out loud, and classic voices read SSML tags out loud.
+        The Voice menu&rsquo;s &ldquo;Default voice&rdquo; is the Text to Speech voice chosen in AntennaHead&rsquo;s Configuration tab.</p>
+        </details>
+        """
+
     /// Listing of the `.txt` files in the configured Text to Speech folder, in
     /// the same name order Listen speaks them chronologically — each with a
     /// checkbox (checked by default) so the user can leave out specific files,

AntennaHead/Services/SDRController.swift

@@ -1819,7 +1819,35 @@ final class SDRController {
             LogStore.shared.log(.error, source: "SDRController", "Text to Speech: nothing to speak")
             return
         }
+        startSpeechSynthPipeline(text: combined, repeatForever: repeatForever, voiceIdentifier: nil, ssml: false, dying: dying)
+    }
+
+    /// Speaks one piece of text typed on the web UI's Text to Speech page
+    /// ("Speak Text" section) once, through the same pipeline as the folder
+    /// Listen. `voiceIdentifier` nil/empty = the Text to Speech voice setting.
+    /// Text starting with `<speak` is passed as SSML (`--ssml`), the same rule
+    /// StationDirector's announcer uses.
+    func speakText(_ text: String, voiceIdentifier: String?) {
+        let trimmed = text.trimmingCharacters(in: .whitespacesAndNewlines)
+        guard !trimmed.isEmpty else {
+            lastError = SDRError.notImplemented("Text to Speech — no text to speak")
+            LogStore.shared.log(.error, source: "SDRController", "Speak Text: nothing to speak")
+            return
+        }
+        let dying = radioTaskPipelineManager.taskItems.compactMap { $0.process }.filter { $0.isRunning }
+        Self.sweepOrphanedHelpers()
+        stopFillerForNewSource()
+        radioTaskPipelineManager.terminate()
+        startSpeechSynthPipeline(text: trimmed, repeatForever: false,
+                                 voiceIdentifier: voiceIdentifier.flatMap { $0.isEmpty ? nil : $0 },
+                                 ssml: trimmed.hasPrefix("<speak"), dying: dying)
+    }
 
+    /// The shared tail of `startTextToSpeech` and `speakText`: stages `text`
+    /// and launches PCMSpeechSynth → sox → PCMUDPSender. The caller has
+    /// already stopped the previous pipeline (`dying` are its processes).
+    private func startSpeechSynthPipeline(text combined: String, repeatForever: Bool,
+                                          voiceIdentifier: String?, ssml: Bool, dying: [Process]) {
         // Stage the combined text in the app's own temp dir (inside the sandbox
         // container, so the PCMSpeechSynth child can read it) rather than passing
         // it as a --text argument, which would risk ARG_MAX for a large folder.
@@ -1850,7 +1878,8 @@ final class SDRController {
         directSamplingQBranch = false
         lastError = nil
 
-        guard let synth = makeSpeechSynthTaskItem(textFileURL: textFileURL, repeatForever: repeatForever),
+        guard let synth = makeSpeechSynthTaskItem(textFileURL: textFileURL, repeatForever: repeatForever,
+                                                  voiceIdentifier: voiceIdentifier, ssml: ssml),
               let resample = makeResampleTaskItem(inputRate: Self.speechSynthSampleRate,
                                                   inputChannels: 1,
                                                   audioOutputFilter: "vol 1"),
@@ -1868,7 +1897,8 @@ final class SDRController {
         launchCurrentPipeline(dying: dying)
     }
 
-    private func makeSpeechSynthTaskItem(textFileURL: URL, repeatForever: Bool) -> TaskItem? {
+    private func makeSpeechSynthTaskItem(textFileURL: URL, repeatForever: Bool,
+                                         voiceIdentifier: String? = nil, ssml: Bool = false) -> TaskItem? {
         let path = helperPath("PCMSpeechSynth")
         guard FileManager.default.isExecutableFile(atPath: path) else {
             lastError = SDRError.notImplemented("PCMSpeechSynth helper missing at \(path)")
@@ -1880,10 +1910,11 @@ final class SDRController {
         item.addArgument("--rate"); item.addArgument(Self.speechSynthSampleRate)
         // Without --voice the helper would use the built-in default (compact
         // Samantha) and ignore the Speech settings entirely.
-        if let voice = SpeechVoicePreference.resolvedIdentifier(
+        if let voice = voiceIdentifier ?? SpeechVoicePreference.resolvedIdentifier(
             overrideKey: SpeechVoicePreference.textToSpeechVoiceKey, sqlite: sqliteController) {
             item.addArgument("--voice"); item.addArgument(voice)
         }
+        if ssml { item.addArgument("--ssml") }
         if repeatForever {
             item.addArgument("--repeat")
             item.addArgument("--gap"); item.addArgument(2)

Web/js/antennahead.js

@@ -2300,6 +2300,36 @@ function textToSpeechListenButtonClicked(form)
 }
 
 
+// Speak Text section (bottom of the Text to Speech page): speaks the text
+// box once with the chosen voice ("" = the Text to Speech voice setting).
+// Text starting with <speak> is SSML — the server decides.
+function speakTextButtonClicked(form)
+{
+  var textArea = form.querySelector("#tts_speak_text");
+  var voiceSelect = form.querySelector("#tts_speak_voice");
+  var payload = {
+    text: textArea ? textArea.value : "",
+    voice: voiceSelect ? voiceSelect.value : ""
+  };
+
+  var getUrl = window.location;
+  var baseUrl = getUrl.protocol + "//" + getUrl.host + "/";
+  var xhttp = new XMLHttpRequest();
+  xhttp.onreadystatechange = function() {
+      if (this.readyState == 4 && this.status == 200) {
+        window.top.nowPlayingTitle = window.document.getElementById("listen_title");
+      }
+    };
+  xhttp.open("POST", baseUrl + "speaktextbuttonclicked.html", true);
+  xhttp.setRequestHeader("Content-Type", "application/json");
+  xhttp.send(JSON.stringify(payload));
+
+  showUpNextInNavBar();
+
+  window.top.postMessage("startaudio", "*");
+}
+
+
 // ---- Play Audio Files (Audio Devices page) ------------------------------
 //
 // Same pattern as Text to Speech above: the folder is chosen in