Add Spanish TTS example with reverb audio processing and Concatenates all audio chunks into a single file .wav per voice.

demo/es.txt

@@ -0,0 +1,8 @@
+--Sí, por Zeus, exclamó.
+--En resumen, ¿estos prisioneros no atribuirán realidad más que a estas sombras?
+--Es inevitable.
+--Supongamos ahora que se les libre de sus cadenas y se les cure de su error; mira lo que
+resultaría naturalmente de la nueva situación en que vamos a colocarlos. Liberamos a uno
+de estos prisioneros. Le obligamos a levantarse, a volver la cabeza, a andar y a mirar hacia
+el lado de la luz: no podrá hacer nada de esto sin sufrir, y el deslumbramiento le impedirá
+distinguir los objetos cuyas sombras antes veía.

examples/es_test_kokoro.py

@@ -0,0 +1,83 @@
+from kokoro import KPipeline
+import soundfile as sf
+import torch
+import numpy as np
+from scipy.signal import fftconvolve
+import os
+
+# Initialize the pipeline for Spanish, auto-select CUDA if available
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+print(f"Using device: {device}")
+pipeline = KPipeline(lang_code='e', device=device)
+
+# Read the text to synthesize from a text file in the demo folder
+with open('../demo/es.txt', 'r', encoding='utf-8') as f:
+    text = f.read()
+
+# List of available Spanish voices
+spanish_voices = ['ef_dora', 'em_alex', 'em_santa']
+
+def simple_reverb(audio, sr=24000, decay=0.08, length=0.07, wet=0.033):
+    # Revberberation effect using a simple impulse response
+    ir_length = int(length * sr)
+    ir = np.logspace(0, -decay, ir_length)
+    ir = ir / np.max(ir)
+    reverbed = fftconvolve(audio, ir, mode='full')[:len(audio)]
+    # Mix the original audio with the reverberated (wet/dry)
+    out = (1 - wet) * audio[:len(reverbed)] + wet * reverbed
+    # Normalize if necessary
+    max_val = np.max(np.abs(out))
+    if max_val > 0:
+        out = out / max_val * 0.95
+    return out
+
+for voice in spanish_voices:
+    print(f"\nTesting voice: {voice}")
+    speed = 0.95
+    split_pattern = r'[.!?]+'
+    audio_chunks = []
+    generator = list(pipeline(
+        text,
+        voice=voice,
+        speed=speed,
+        split_pattern=split_pattern,
+    ))
+    fade_duration = 0.5  # seconds
+    fade_samples = int(fade_duration * 24000)
+    minimum_pause = 0  # seconds
+    initial_silence = 0.38  # seconds
+    initial_silence_samples = int(initial_silence * 24000)
+
+    for i, (gs, ps, audio) in enumerate(generator):
+        print(f"Sample {i}")
+        print("Graphemes:", gs)
+        print("Phonemes:", ps)
+        audio_np = audio.cpu().numpy() if hasattr(audio, 'cpu') else np.array(audio)
+        if len(audio_np) > fade_samples:
+            if i == 0:
+                # Add silence at the beginning and apply fade-in to the combined block
+                silence = np.zeros(initial_silence_samples, dtype=np.float32)
+                audio_with_silence = np.concatenate([silence, audio_np])
+                fade_in = np.linspace(0, 1, fade_samples)
+                audio_with_silence[:fade_samples] *= fade_in
+                # Delete the initial silence after fade-in to avoid delaying the phrase
+                audio_np = audio_with_silence[initial_silence_samples:]
+            else:
+                fade_in = np.linspace(0, 1, fade_samples)
+                audio_np[:fade_samples] *= fade_in
+            if i < len(generator) - 1:
+                fade_out = np.linspace(1, 0, fade_samples)
+                audio_np[-fade_samples:] *= fade_out
+        audio_chunks.append(audio_np)
+        if i < len(generator) - 1:
+            pause = np.zeros(int(minimum_pause * 24000), dtype=np.float32)
+            audio_chunks.append(pause)
+    if audio_chunks:
+        full_audio = np.concatenate(audio_chunks)
+        # Apply subtle reverb
+        full_audio = simple_reverb(full_audio, sr=24000)
+        export_dir = 'audio_exports'
+        os.makedirs(export_dir, exist_ok=True)
+        filename = os.path.join(export_dir, f'output_{voice}.wav')
+        sf.write(filename, full_audio, 24000)
+        print(f"Audio saved to {filename}")