Update app.py
Browse files
app.py
CHANGED
|
@@ -1,42 +1,64 @@
|
|
|
|
|
| 1 |
import numpy as np
|
| 2 |
import librosa
|
| 3 |
import pyworld as pw
|
| 4 |
import soundfile as sf
|
| 5 |
-
from transformers import AutoProcessor, AutoModel
|
| 6 |
import gradio as gr
|
|
|
|
| 7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
|
| 9 |
-
from transformers import pipeline
|
| 10 |
-
|
| 11 |
-
# 軽量TTS
|
| 12 |
-
tts = pipeline("text-to-speech", model="p0p0pp/KoeTTS")
|
| 13 |
|
|
|
|
|
|
|
|
|
|
| 14 |
def tts_standard(text):
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 19 |
|
| 20 |
|
| 21 |
# ==========================
|
| 22 |
-
# 2) F0
|
| 23 |
# ==========================
|
| 24 |
def convert_to_kansai_pitch(audio, sr):
|
| 25 |
-
"""
|
|
|
|
|
|
|
| 26 |
|
| 27 |
# WORLDで分解
|
| 28 |
f0, sp, ap = pw.wav2world(audio.astype(np.float64), sr)
|
| 29 |
|
| 30 |
-
#
|
| 31 |
nonzero_idx = np.where(f0 > 0)[0]
|
| 32 |
if len(nonzero_idx) < 2:
|
| 33 |
return audio
|
| 34 |
|
| 35 |
-
# 基本
|
| 36 |
high = np.max(f0[nonzero_idx])
|
| 37 |
low = np.min(f0[nonzero_idx])
|
| 38 |
|
| 39 |
-
# HLパターン
|
| 40 |
hl_curve = np.linspace(high, low, len(f0))
|
| 41 |
f0_new = hl_curve
|
| 42 |
|
|
@@ -46,7 +68,7 @@ def convert_to_kansai_pitch(audio, sr):
|
|
| 46 |
|
| 47 |
|
| 48 |
# ==========================
|
| 49 |
-
# 3) 統合
|
| 50 |
# ==========================
|
| 51 |
def kansai_tts(text):
|
| 52 |
sr, audio_std = tts_standard(text)
|
|
@@ -58,8 +80,13 @@ def kansai_tts(text):
|
|
| 58 |
# Gradio UI
|
| 59 |
# ==========================
|
| 60 |
with gr.Blocks() as demo:
|
| 61 |
-
gr.Markdown("## 🔊 Kansign — Kansai Accent TTS(
|
| 62 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
audio_out = gr.Audio(label="関西イントネーション音声", type="numpy")
|
| 64 |
|
| 65 |
btn = gr.Button("関西イントネーションで喋る")
|
|
|
|
| 1 |
+
import os
|
| 2 |
import numpy as np
|
| 3 |
import librosa
|
| 4 |
import pyworld as pw
|
| 5 |
import soundfile as sf
|
|
|
|
| 6 |
import gradio as gr
|
| 7 |
+
from openai import OpenAI
|
| 8 |
|
| 9 |
+
# -----------------------------
|
| 10 |
+
# OpenAI Client
|
| 11 |
+
# -----------------------------
|
| 12 |
+
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
| 13 |
+
client = OpenAI(api_key=OPENAI_API_KEY)
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
+
# ==========================
|
| 17 |
+
# 1) OpenAI 標準音声 TTS
|
| 18 |
+
# ==========================
|
| 19 |
def tts_standard(text):
|
| 20 |
+
"""
|
| 21 |
+
OpenAI gpt-4o-mini-ttsで基本音声を生成
|
| 22 |
+
"""
|
| 23 |
+
response = client.audio.speech.create(
|
| 24 |
+
model="gpt-4o-mini-tts",
|
| 25 |
+
voice="alloy",
|
| 26 |
+
input=text
|
| 27 |
+
)
|
| 28 |
+
audio_bytes = response.read()
|
| 29 |
+
|
| 30 |
+
# WAVとして読み込む
|
| 31 |
+
sr, audio = sf.read(
|
| 32 |
+
sf.SoundFile(
|
| 33 |
+
io.BytesIO(audio_bytes),
|
| 34 |
+
mode='r'
|
| 35 |
+
)
|
| 36 |
+
)
|
| 37 |
+
|
| 38 |
+
return sr, audio.astype(np.float32)
|
| 39 |
|
| 40 |
|
| 41 |
# ==========================
|
| 42 |
+
# 2) F0を関西イントネーションHL型に変換
|
| 43 |
# ==========================
|
| 44 |
def convert_to_kansai_pitch(audio, sr):
|
| 45 |
+
"""
|
| 46 |
+
標準TTS音声 → 関西弁イントネーション(HLパターン)へ変換
|
| 47 |
+
"""
|
| 48 |
|
| 49 |
# WORLDで分解
|
| 50 |
f0, sp, ap = pw.wav2world(audio.astype(np.float64), sr)
|
| 51 |
|
| 52 |
+
# 有効フレーム
|
| 53 |
nonzero_idx = np.where(f0 > 0)[0]
|
| 54 |
if len(nonzero_idx) < 2:
|
| 55 |
return audio
|
| 56 |
|
| 57 |
+
# 基本周波数の範囲
|
| 58 |
high = np.max(f0[nonzero_idx])
|
| 59 |
low = np.min(f0[nonzero_idx])
|
| 60 |
|
| 61 |
+
# HLパターン(前半高 → 後半低)
|
| 62 |
hl_curve = np.linspace(high, low, len(f0))
|
| 63 |
f0_new = hl_curve
|
| 64 |
|
|
|
|
| 68 |
|
| 69 |
|
| 70 |
# ==========================
|
| 71 |
+
# 3) 統合:Kansai TTS
|
| 72 |
# ==========================
|
| 73 |
def kansai_tts(text):
|
| 74 |
sr, audio_std = tts_standard(text)
|
|
|
|
| 80 |
# Gradio UI
|
| 81 |
# ==========================
|
| 82 |
with gr.Blocks() as demo:
|
| 83 |
+
gr.Markdown("## 🔊 Kansign — Kansai Accent TTS(OpenAI版・安定稼働)")
|
| 84 |
+
|
| 85 |
+
text_in = gr.Textbox(
|
| 86 |
+
label="テキストを入力(例:なんでやねん)",
|
| 87 |
+
value="なんでやねん"
|
| 88 |
+
)
|
| 89 |
+
|
| 90 |
audio_out = gr.Audio(label="関西イントネーション音声", type="numpy")
|
| 91 |
|
| 92 |
btn = gr.Button("関西イントネーションで喋る")
|