Narration: all 28 English Kokoro voices, British pronunciation for the British ones

README.md

@@ -55,7 +55,7 @@ models:
 
 **Transcribe tab (speech to text, speech recognition, automatic speech recognition):** Whisper large v3 turbo (openai) writes down speech from a file or the microphone, in its own language or translated to English.
 
-**Narration tab (text to speech, speech generation, machine speech):** Kokoro (hexgrad) reads your text aloud in a choice of eight voices.
+**Narration tab (text to speech, speech generation, machine speech):** Kokoro (hexgrad) reads your text aloud in a choice of 28 English voices, American and British.
 
 **Describe tab:** three ways to describe what you have.
 

app.py

@@ -82,9 +82,18 @@ LISTEN_MODEL = "laion/clap-htsat-unfused"      # Recognises genre, mood, instrum
 SPEECH_MODEL = "openai/whisper-large-v3-turbo" # Speech to text
 VOICE = "af_heart"                             # Kokoro voice
 VOICE_RATE = 24000
+# All 28 English voices of Kokoro. The first eight keep the names they had when there were only eight.
 VOICES = {"Heart (woman)": "af_heart", "Bella (woman)": "af_bella", "Nicole (woman, soft)": "af_nicole",
           "Sarah (woman)": "af_sarah", "Adam (man)": "am_adam", "Michael (man)": "am_michael",
-          "Fenrir (man, deep)": "am_fenrir", "Puck (man, lively)": "am_puck"}
+          "Fenrir (man, deep)": "am_fenrir", "Puck (man, lively)": "am_puck",
+          "Alloy (woman)": "af_alloy", "Aoede (woman)": "af_aoede", "Jessica (woman)": "af_jessica",
+          "Kore (woman)": "af_kore", "Nova (woman)": "af_nova", "River (woman)": "af_river", "Sky (woman)": "af_sky",
+          "Echo (man)": "am_echo", "Eric (man)": "am_eric", "Liam (man)": "am_liam", "Onyx (man)": "am_onyx",
+          "Santa (man)": "am_santa",
+          "Alice (British woman)": "bf_alice", "Emma (British woman)": "bf_emma",
+          "Isabella (British woman)": "bf_isabella", "Lily (British woman)": "bf_lily",
+          "Daniel (British man)": "bm_daniel", "Fable (British man)": "bm_fable",
+          "George (British man)": "bm_george", "Lewis (British man)": "bm_lewis"}
 MAX_NARRATION_CHARS = 3000
 WORDS_PER_SECOND = 2.5
 

tts_worker.py

@@ -14,13 +14,26 @@ os.dup2(2, 1)
 import numpy as np
 from kokoro import KPipeline
 
-pipe = KPipeline(lang_code="a", device="cpu")     # Kokoro is tiny (82M) and fast enough on the CPU
+pipes = {"a": KPipeline(lang_code="a", device="cpu")}     # Kokoro is tiny (82M) and fast enough on the CPU
+
+
+def pipe_for(voice):
+    """American pronunciation for the af_/am_ voices, British for bf_/bm_. Both share one copy of the model."""
+    code = "b" if voice.startswith("b") else "a"
+    if code not in pipes:
+        try:
+            pipes[code] = KPipeline(lang_code=code, model=pipes["a"].model)
+        except Exception as e:
+            print(f"British pronunciation unavailable, using American: {type(e).__name__}: {e}", file=sys.stderr)
+            pipes[code] = pipes["a"]
+    return pipes[code]
+
 
 for line in sys.stdin:
     try:
         job = json.loads(line)
         chunks = [np.asarray(audio, dtype=np.float32)
-                  for _, _, audio in pipe(job["text"], voice=job["voice"], speed=float(job["speed"]))]
+                  for _, _, audio in pipe_for(job["voice"])(job["text"], voice=job["voice"], speed=float(job["speed"]))]
         np.save(job["out"], np.concatenate(chunks) if chunks else np.zeros(0, dtype=np.float32))
         answer = {"ok": True}
     except Exception as e: