first draft implementation

.vscode/launch.json

@@ -0,0 +1,16 @@
+{
+    // Use IntelliSense to learn about possible attributes.
+    // Hover to view descriptions of existing attributes.
+    // For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Streamlit: Run and Debug",
+            "type": "debugpy",
+            "request": "launch",
+            "program": "../../venv/bin/streamlit",
+            "console": "integratedTerminal",
+            "args": ["run", "app.py"]
+        }
+    ]
+}

app.py

@@ -1,4 +1,108 @@
+import time
+
 import streamlit as st
 
-x = st.slider('Select a value')
-st.write(x, 'squared is', x * x)
+from zerovox.tts.synthesize import ZeroVoxTTS
+from zerovox.g2p.g2p import DEFAULT_G2P_MODEL_NAME_DE, DEFAULT_G2P_MODEL_NAME_EN
+
+SAMPLE_RATE=24000 # FIXME
+TTS_MODEL_NAME='tts_en_de_zerovox_alpha1'
+
+if "text" not in st.session_state:
+    st.session_state.text = "Welcome to the world of speech synthesis!"
+
+if "message" not in st.session_state:
+    st.session_state.message = "READY."
+
+def update_text_input():
+    global text
+    if st.session_state['lang'] == "en":
+        st.session_state.text = "Welcome to the world of speech synthesis!"
+    elif st.session_state['lang'] == "de":
+        st.session_state.text = "Willkommen in der Welt der Sprachsynthese!"
+    text = st.session_state.text
+
+def do_synth():
+
+    global lang, status, speakerref, text, playback, meldec
+
+
+    synth = None
+    if 'synth' in st.session_state:
+        synth = st.session_state.synth
+
+        if synth.language != lang or synth.meldec_model != st.session_state['meldec']:
+            synth = None # trigger reload
+
+    if not synth:
+
+        status.update(label="loading the model...", state="running")
+
+        g2p_model = DEFAULT_G2P_MODEL_NAME_DE if lang=='de' else DEFAULT_G2P_MODEL_NAME_EN
+
+        st.session_state.modelcfg, st.session_state.synth = ZeroVoxTTS.load_model(TTS_MODEL_NAME,
+                                                                                  g2p=g2p_model,
+                                                                                  lang=lang,
+                                                                                  meldec_model=st.session_state['meldec'],
+                                                                                  infer_device='cpu',
+                                                                                  num_threads=-1,
+                                                                                  do_compile=False,
+                                                                                  verbose=False)
+
+    synth = st.session_state.synth
+    modelcfg = st.session_state.modelcfg
+
+    status.update(label="computing speaker embedding...", state="running")
+
+    spkemb = synth.speaker_embed(ZeroVoxTTS.get_speakerref(speakerref, modelcfg['audio']['sampling_rate']))    
+
+    status.update(label="synthesizing...", state="running")
+
+    start_time = time.time()
+
+    wav, phoneme, length = synth.tts(text, spkemb)
+
+    elapsed_time = time.time() - start_time
+
+    message = f"synth time: {elapsed_time:.2f} sec"
+    wav_len = wav.shape[0] / modelcfg['audio']['sampling_rate']
+    message += f", voice length: {wav_len:.2f} sec"
+    real_time_factor = wav_len / elapsed_time
+    message += f", rtf: {real_time_factor:.2f}"
+
+    st.session_state.message = message
+
+    status.update(label=message, state="complete")
+
+    st.session_state.wav = wav
+    playback.audio(wav, sample_rate=SAMPLE_RATE)
+
+    
+
+lang = st.selectbox("Language",
+                    ["en", "de"],
+                    on_change=update_text_input,
+                    key='lang')
+
+meldec = st.selectbox("MEL decoder",
+                      ["meldec-libritts-multi-band-melgan-v2", "meldec-libritts-hifigan-v1"],
+                      #on_change=update_text_input,
+                      key='meldec')
+
+speakerref = st.selectbox("Voice sample", ZeroVoxTTS.available_speakerrefs())
+
+st.audio(ZeroVoxTTS.get_speakerref(speakerref, SAMPLE_RATE), sample_rate=SAMPLE_RATE)
+
+text = st.text_input("Text to synthesize", value=st.session_state.text)
+
+st.button("Synthesize!", type="primary", on_click=do_synth)
+
+status = st.status(st.session_state.message, state="complete")
+
+if 'wav' in st.session_state:
+
+    playback = st.audio(st.session_state.wav, sample_rate=SAMPLE_RATE)
+
+else:
+
+    playback = st.empty()