Fix Fish-Speech 0-voices bug, add inline emotion-tag reference to Try It Out (v1.20.8)

Fish-Speech clones from the app's own saved WAV library but was missing from
the set of backends populated with those voices, so the UI probed the Fish
server for a voice-listing endpoint it does not have, showed "Fetched 0
voices" and left the engine unusable despite being healthy. It now reports
the full library (218 voices).

Also adds a Fish-only panel listing all 49 documented emotion tags with
click-to-insert and a multi-emotion example, and corrects the backend's
advertised capabilities: it was flagged style_aware with "emotion markers are
honoured per request", but measurement shows a reference clip's in-context
prosody overwhelms inline tags (10x loudness spread across emotions without a
reference, 1.2x with one). It is now described by what it does well:
deterministic, byte-reproducible cloning.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

CHANGELOG.md

@@ -5,6 +5,16 @@ Follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · versioned wi
 
 ---
 
+## [1.20.8] — 2026-08-12
+
+### Added
+- **Fish Audio inline emotion-tag reference in Try It Out.** Selecting a Fish backend reveals a panel with all 49 documented tags (24 basic + 25 advanced); click any tag to insert it at the cursor, plus an "Insert example" button that loads a multi-emotion sample. Shown only for Fish, since it is the only engine that reads bracket tags from the text itself — other engines would speak "[happy]" aloud as literal words.
+
+### Fixed
+- **Fish-Speech showed "Fetched 0 voices" and was unusable from the UI.** It clones from the app's own saved WAV library, but was missing from the set of backends populated with those voices, so the code probed the Fish server for a voice-listing endpoint it does not have and got nothing. Now reports the full library (218 voices).
+- **Renamed the backend to "Fish Audio S2 Pro (Clone)"** — verified the running checkpoint is `fishaudio/s2-pro` at revision `1de9996b`, matching upstream HuggingFace exactly.
+- **Corrected Fish's advertised capabilities.** It was listed as `style_aware: true` with "Strong: emotion/tone markers … are honoured per request" — measured to be wrong while cloning. With a reference clip, S2-Pro prepends it as an in-context example and imitates its prosody, which overwhelms inline tags: identical text showed a 10x loudness spread across emotions with no reference, but only 1.2x with one ([screaming] vs [whisper] was inaudible). Every lever was tested — canonical tags, free-form descriptions, repeated and mid-sentence tags, temperature to 1.0, truncated references, `reference_id` vs inline. The backend is now billed on what it does well: deterministic, byte-reproducible cloning.
+
 ## [1.20.6] — 2026-08-11
 
 ### Fixed

VERSION

@@ -1 +1 @@
-1.20.6
+1.20.8

routes/tts.py

@@ -80,7 +80,7 @@ def _backend_display_name(backend: str, url: str) -> str:
         "kokoro":         "Kokoro FastAPI (82M)",
         "vibevoice":      "VibeVoice TTS",
         "xtts":           "XTTS v2",
-        "fishspeech":     "Fish-Speech (Clone + Emotion)",
+        "fishspeech":     "Fish Audio S2 Pro (Clone)",
     }
     port = _backend_port_label(url)
     return f"{port} {names.get(backend, backend)}" if port else names.get(backend, backend)
@@ -168,12 +168,22 @@ def _backend_capabilities(backend: str) -> dict:
             "uses_wav": False, "style_aware": False, "true_streaming": False,
             "speed": "~0.3× GPU", "latency": "1–3 s", "quality": "High", "ram": "~5 GB",
         },
+        # Measured live against this exact checkpoint (fishaudio/s2-pro, rev 1de9996b):
+        # inline [tag] emotion markers work well ONLY when no reference audio is sent.
+        # With a reference (i.e. whenever cloning a library voice) S2-Pro prepends the
+        # clip as an in-context example and imitates its prosody, which overwhelms the
+        # tags: identical text scored a 10× loudness spread across emotions with no
+        # reference, but only 1.2× with one — [screaming] vs [whisper] was inaudible.
+        # Every lever was tried (canonical tags, free-form descriptions, repeated and
+        # mid-sentence tags, temperature up to 1.0, 3s truncated references,
+        # reference_id vs inline). So this backend is billed on what it actually does
+        # well: rock-solid deterministic cloning, not per-line emotion.
         "fishspeech": {
-            "purpose": "Clone a saved WAV voice AND control per-line tone via inline emotion markers.",
-            "identity": "Clones the selected voice's reference WAV — consistent speaker identity across lines.",
-            "style": "Strong: emotion/tone markers like (angry), (whispering), (excited) are honoured per request.",
-            "best_for": "Consistent character voices that still react to tone changes — ideal for the Rehearser.",
-            "uses_wav": True, "style_aware": True, "true_streaming": False,
+            "purpose": "Clone a saved WAV voice with byte-identical, reproducible output.",
+            "identity": "Excellent — clones the selected voice's reference WAV and is fully deterministic (same input always yields the same audio).",
+            "style": "Weak while cloning: inline [tag] markers are overwhelmed by the reference clip's own delivery. Tone follows the reference, so the emotion of the reference recording is what you hear.",
+            "best_for": "Consistent character voices and resumable bulk exports, where identical re-runs matter more than per-line emotion.",
+            "uses_wav": True, "style_aware": False, "true_streaming": False,
             "speed": "~0.4× GPU", "latency": "1–4 s", "quality": "Premium", "ram": "~4 GB",
         },
     }
@@ -369,7 +379,12 @@ def _active_library_voice_options(settings: dict) -> list[dict]:
 
 def _fetch_backend_voices(settings: dict, backend: str) -> list:
     backend = _clean_preview_backend(backend)
-    if backend in {"voice_clone", "streaming", "nvidia_zeroshot", "nvidia_flow"}:
+    # Fish-Speech clones from the app's own saved WAV library (see
+    # _fishspeech_request_audio, which resolves the voice via _find_voice_audio) and
+    # exposes no voice-listing endpoint of its own. Without it in this set the
+    # probe below found nothing and reported voice_count 0, so the Engines page
+    # listed it with no selectable voices even while the server was healthy.
+    if backend in {"voice_clone", "streaming", "nvidia_zeroshot", "nvidia_flow", "fishspeech"}:
         return _active_library_voice_options(settings)
     tts_url = _validate_http_url(_preview_backend_base_url(settings, backend), allow_private=True).rstrip("/")
     key = (settings.get("voice_design_api_key") if backend == "voice_design" else settings.get("tts_api_key")) or ""

static/index.html

@@ -10,7 +10,7 @@
   <meta name="format-detection" content="telephone=no">
   <meta name="color-scheme" content="light dark">
   <meta name="theme-color" content="#2563EB">
-  <meta name="app-version" content="1.20.6">
+  <meta name="app-version" content="1.20.8">
   <link rel="manifest" href="/manifest.webmanifest">
   <link rel="icon" href="/static/icon.svg" type="image/svg+xml">
   <link rel="apple-touch-icon" href="/static/icon.svg">
@@ -27,7 +27,7 @@
 
   <!-- ── Core styles (local — no CDN dependency for first paint) ────────── -->
   <link rel="stylesheet" href="/static/vendor/mdi/materialdesignicons.min.css">
-  <link rel="stylesheet" href="/static/style.css?v=1.20.6">
+  <link rel="stylesheet" href="/static/style.css?v=1.20.8">
 
 
   <!-- ── Flag icons — non-blocking (loaded async, icons appear after JS) ── -->
@@ -378,7 +378,7 @@ window.toggleNavTree = function(treeId, chevronId) {
 </script>
 
 <!-- loader.js: fetches sections → loads JS modules → removes skeleton -->
-<script src="/static/loader.js?v=1.20.6"></script>
+<script src="/static/loader.js?v=1.20.8"></script>
 
 </body>
 </html>

static/js/rehearser.js

@@ -1903,7 +1903,21 @@ function _buildInstruct(voiceProfile, emotion, voiceId) {
   // through. Emotion is the one thing that MUST vary per line; accent is a
   // constant reminder the voice's own identity should mostly already carry,
   // so it goes last, not first.
-  const parts = [e ? emotionTmpl(effectiveEmotion) : '', p, accent].filter(Boolean);
+  // ...and when an emotion IS present, the accent clause is dropped entirely rather
+  // than merely demoted. Measured on the deterministic voice_clone backend (identical
+  // input reproduces bit-identical audio, so these differences are real, not sampling
+  // noise): with an emotion-only instruct the tones separate cleanly and in the right
+  // order — whisper 206Hz, sad 170Hz, neutral 225Hz, happy 234Hz, angry 233Hz, scared
+  // 239Hz, with sad also slower and quieter. Appending "Speak with an authentic native
+  // German accent." to those same instructs collapsed the whole range to 222-250Hz and
+  // erased sad almost completely. Two style directives in one instruct compete, and the
+  // trailing one wins — which is exactly the failure Qwen3-TTS's own prompting guidance
+  // warns about for conflicting attributes. A cloned voice already carries its accent
+  // from the reference WAV, so accent reinforcement is the cheaper of the two to lose;
+  // lines with no emotion still get it.
+  const parts = e
+    ? [emotionTmpl(effectiveEmotion), p].filter(Boolean)
+    : [p, accent].filter(Boolean);
   return parts.join(' ');
 }
 

static/js/tts-preview.js

@@ -213,6 +213,85 @@ $('fetch-tts-voices-btn').addEventListener('click', async () => {
 // while style-aware backends (VoiceDesign/CustomVoice) keep working through
 // the instruct field exactly as before.
 function _ttsIsFishBackend(id) { return /fish/i.test(id || ''); }
+
+// ── Fish Audio inline [tag] reference ───────────────────────────────────────
+// Fish is the only backend here that reads bracket tags straight out of the text
+// (every other engine takes a separate instruct field and would speak "[happy]"
+// aloud as literal words), so this panel only appears for Fish backends.
+// Tag list per Fish Audio's emotion documentation: 24 basic + 25 advanced.
+const FISH_EMOTION_TAGS = {
+  'Basic emotions': [
+    ['happy','Cheerful, upbeat'], ['sad','Melancholic, downcast'], ['angry','Frustrated, aggressive'],
+    ['excited','Energetic, enthusiastic'], ['calm','Peaceful, relaxed'], ['nervous','Anxious, uncertain'],
+    ['confident','Assertive, self-assured'], ['surprised','Shocked, amazed'], ['satisfied','Content, pleased'],
+    ['delighted','Very pleased, joyful'], ['scared','Frightened, fearful'], ['worried','Concerned, troubled'],
+    ['upset','Disturbed, distressed'], ['frustrated','Annoyed, exasperated'], ['depressed','Very sad, hopeless'],
+    ['empathetic','Understanding, caring'], ['embarrassed','Ashamed, awkward'], ['disgusted','Repelled, revolted'],
+    ['moved','Emotionally touched'], ['proud','Accomplished, satisfied'], ['relaxed','At ease, casual'],
+    ['grateful','Thankful, appreciative'], ['curious','Inquisitive, interested'], ['sarcastic','Ironic, mocking'],
+  ],
+  'Advanced emotions': [
+    ['disdainful','Contemptuous, scornful'], ['unhappy','Discontent, dissatisfied'], ['anxious','Very worried, uneasy'],
+    ['hysterical','Uncontrollably emotional'], ['indifferent','Uncaring, neutral'], ['uncertain','Doubtful, unsure'],
+    ['doubtful','Skeptical, questioning'], ['confused','Puzzled, perplexed'], ['disappointed','Let down'],
+    ['regretful','Sorry, remorseful'], ['guilty','Culpable, responsible'], ['ashamed','Deeply embarrassed'],
+    ['jealous','Envious, resentful'], ['envious','Wanting what others have'], ['hopeful','Optimistic about future'],
+    ['optimistic','Positive outlook'], ['pessimistic','Negative outlook'], ['nostalgic','Longing for the past'],
+    ['lonely','Isolated, alone'], ['bored','Uninterested, weary'], ['contemptuous','Showing contempt'],
+    ['sympathetic','Showing sympathy'], ['compassionate','Showing deep care'], ['determined','Resolved, decided'],
+    ['resigned','Accepting defeat'],
+  ],
+};
+const FISH_TAG_SAMPLE_TEXT = `[happy] I got the promotion!
+[uncertain] But... it means relocating.
+[sad] I'll miss everyone here.
+[hopeful] Though it's a great opportunity.
+[determined] I'm going to make it work!`;
+
+function _fishInsertTag(tag) {
+  const ta = $('preview-text-area');
+  if (!ta) return;
+  const snippet = `[${tag}] `;
+  const s = ta.selectionStart ?? ta.value.length, e = ta.selectionEnd ?? ta.value.length;
+  ta.value = ta.value.slice(0, s) + snippet + ta.value.slice(e);
+  const pos = s + snippet.length;
+  ta.setSelectionRange(pos, pos);
+  ta.focus();
+}
+
+function _fishTagsRender() {
+  const body = $('fish-tags-body');
+  if (!body || body.dataset.built) return;
+  body.innerHTML = Object.entries(FISH_EMOTION_TAGS).map(([group, tags]) => `
+    <div class="fish-tags-group">
+      <div class="fish-tags-group-title">${escHtml(group)} <span class="note">(${tags.length})</span></div>
+      <div class="fish-tags-grid">
+        ${tags.map(([t, d]) => `<button type="button" class="fish-tag" data-tag="${escHtml(t)}" title="${escHtml(d)}">[${escHtml(t)}]</button>`).join('')}
+      </div>
+    </div>`).join('');
+  body.querySelectorAll('.fish-tag').forEach(b =>
+    b.addEventListener('click', () => _fishInsertTag(b.dataset.tag)));
+  body.dataset.built = '1';
+}
+
+function _fishTagsSync() {
+  const panel = $('fish-tags-panel');
+  if (!panel) return;
+  panel.hidden = !_ttsIsFishBackend($('tts-backend-select')?.value || '');
+}
+
+(function initFishTagsPanel() {
+  const toggle = $('fish-tags-toggle'), body = $('fish-tags-body');
+  toggle?.addEventListener('click', () => {
+    _fishTagsRender();
+    body.hidden = !body.hidden;
+    toggle.textContent = body.hidden ? 'Show tags' : 'Hide tags';
+  });
+  $('fish-tags-sample')?.addEventListener('click', () => {
+    const ta = $('preview-text-area');
+    if (ta) { ta.value = FISH_TAG_SAMPLE_TEXT; ta.focus(); }
+  });
+})();
 function _ttsApplyEmotionTag(text, emotionValue) {
   const backend = $('tts-backend-select')?.value || '';
   if (!_ttsIsFishBackend(backend) || !emotionValue) return text;
@@ -250,6 +329,9 @@ setTimeout(function initPreviewEmotionPicker() {
       if (styleInput && sel.value) styleInput.value = sel.value;
     }
   });
+  // Sections are injected asynchronously, so sync the Fish panel once the
+  // backend <select> actually exists rather than at module-eval time.
+  _fishTagsSync();
 }, 0);
 
 $('tts-backend-select').addEventListener('change', () => {
@@ -261,6 +343,7 @@ $('tts-backend-select').addEventListener('change', () => {
   $('save-preview-mp3-btn').disabled = true;
   $('save-preview-btn').disabled = true;
   $('preview-emotion-select')?.dispatchEvent(new Event('change'));
+  _fishTagsSync();
 });
 
 $('tts-voice-select').addEventListener('change', updatePreviewVoiceMatchPanel);

static/sections/s-tryout.html

@@ -71,6 +71,26 @@
             </select>
             <span id="preview-emotion-help" class="note" style="display:none;margin-top:4px"></span>
           </div>
+
+          <!-- Fish Audio inline [tag] reference — only shown when a Fish backend is picked,
+               since it is the only engine that reads bracket tags out of the text itself. -->
+          <div class="field fish-tags-panel" id="fish-tags-panel" hidden style="margin-top:10px">
+            <div class="fish-tags-head">
+              <span><span class="mdi mdi-fish"></span> Fish Audio inline emotion tags</span>
+              <span style="flex:1"></span>
+              <button type="button" class="btn-secondary btn-sm" id="fish-tags-sample">Insert example</button>
+              <button type="button" class="btn-secondary btn-sm" id="fish-tags-toggle">Show tags</button>
+            </div>
+            <div class="note" style="margin-top:6px">
+              Place a tag in square brackets immediately before the words it should affect; it applies until the next tag.
+              Click any tag to insert it at the cursor.
+            </div>
+            <div id="fish-tags-warn" class="note" style="color:var(--yellow);margin-top:6px">
+              <span class="mdi mdi-alert-outline"></span> Measured on this server: while <strong>cloning a saved voice</strong> these tags
+              have little audible effect — the reference clip's own delivery dominates. They work strongly only when no reference voice is used.
+            </div>
+            <div id="fish-tags-body" hidden style="margin-top:8px"></div>
+          </div>
           <div class="field" style="margin-top:10px">
             <label style="display:flex;align-items:center;gap:8px">
               Style instruction

static/style.css

@@ -7291,3 +7291,30 @@ code { background: var(--panel); border-radius: 4px; padding: 1px 5px; font-fami
 .reader-word.stat-ready   { background: rgba(34,197,94,.14); }
 .reader-word.stat-reading { background: rgba(37,99,235,.15); }
 .reader-word.is-word { background: rgba(37,99,235,.42) !important; box-shadow: 0 0 0 1px rgba(37,99,235,.6); }
+
+/* ── Fish Audio inline emotion-tag reference (Try It Out) ─────────────────── */
+/* Uses the app's real theme tokens (--surface/--panel/--border/--text), not
+   hardcoded colours, so the panel follows both the light and dark themes. */
+.fish-tags-panel {
+  border: 1px solid var(--border);
+  border-radius: var(--radius);
+  padding: 10px 12px;
+  background: var(--panel);
+}
+.fish-tags-head { display: flex; align-items: center; gap: 8px; font-weight: 600; color: var(--text); }
+.fish-tags-group { margin-top: 10px; }
+.fish-tags-group-title { font-size: 12px; font-weight: 600; color: var(--subtext); margin-bottom: 6px; }
+.fish-tags-grid { display: flex; flex-wrap: wrap; gap: 6px; }
+.fish-tag {
+  font-family: ui-monospace, SFMono-Regular, Menlo, monospace;
+  font-size: 12px;
+  padding: 3px 8px;
+  border-radius: 5px;
+  border: 1px solid var(--border);
+  background: var(--surface);
+  color: var(--text);
+  cursor: pointer;
+  transition: background .12s, border-color .12s, color .12s;
+}
+.fish-tag:hover { background: var(--accent); border-color: var(--accent); color: var(--on-accent); }
+.fish-tag:focus-visible { outline: 2px solid var(--accent); outline-offset: 1px; }