SPEAK / app.py
Benitha's picture
Upload app.py
c2962bc verified
import gradio as gr
from transformers import pipeline
import whisper
from collections import Counter
import matplotlib.pyplot as plt
# Load models
emotion_classifier = pipeline("audio-classification", model="superb/hubert-large-superb-er")
whisper_model = whisper.load_model("base")
def create_emotion_chart(labels, scores):
emoji_map = {
"hap": "๐Ÿ˜Š Happy", "sad": "๐Ÿ˜” Sad", "neu": "๐Ÿ˜ Neutral",
"ang": "๐Ÿ˜  Angry", "fea": "๐Ÿ˜จ Fear", "dis": "๐Ÿคข Disgust", "sur": "๐Ÿ˜ฎ Surprise"
}
color_map = {
"hap": "#facc15", "sad": "#60a5fa", "neu": "#a1a1aa",
"ang": "#ef4444", "fea": "#818cf8", "dis": "#14b8a6", "sur": "#f472b6"
}
display_labels = [emoji_map.get(label, label) for label in labels]
colors = [color_map.get(label, "#60a5fa") for label in labels]
fig, ax = plt.subplots(figsize=(5, 3.5))
bars = ax.barh(display_labels, scores, color=colors, edgecolor="black", height=0.5)
for bar, score in zip(bars, scores):
ax.text(bar.get_width() + 0.02, bar.get_y() + bar.get_height() / 2, f"{score:.2f}", va='center', fontsize=10)
ax.set_xlim(0, 1)
ax.set_title("๐ŸŽญ Emotion Confidence Scores", fontsize=13, pad=10)
ax.invert_yaxis()
ax.set_facecolor("#f9fafb")
fig.patch.set_facecolor("#f9fafb")
for spine in ax.spines.values():
spine.set_visible(False)
ax.tick_params(axis='x', colors='gray')
ax.tick_params(axis='y', colors='gray')
return fig
def generate_next_moves(dominant_emotion, conf_score, transcript=""):
suggestions = []
harsh_words = ["bad", "ugly", "terrible", "hate", "worst"]
positive_tone_negative_words = any(word in transcript.lower() for word in harsh_words) if "happiness" in dominant_emotion else False
if 'sadness' in dominant_emotion:
suggestions.append("Your tone feels low โ€” try lifting the pitch slightly to bring more warmth.")
suggestions.append("Even if the words are positive, a brighter tone helps convey enthusiasm.")
elif 'happiness' in dominant_emotion and conf_score >= 80:
suggestions.append("Nice energy! Try modulating your tone even more for emphasis in key moments.")
suggestions.append("Experiment with subtle emotional shifts as you speak for more depth.")
elif 'neutral' in dominant_emotion:
suggestions.append("Add inflection to break a monotone pattern โ€” especially at the ends of sentences.")
suggestions.append("Highlight your message by stressing emotionally important words.")
elif conf_score < 50:
suggestions.append("Try exaggerating vocal ups and downs when reading to unlock more expression.")
suggestions.append("Slow down slightly and stretch certain words to vary your delivery.")
else:
suggestions.append("Keep practicing tone variation โ€” youโ€™re building a solid base.")
if positive_tone_negative_words:
suggestions.append("Your tone was upbeat, but the word choices were harsh โ€” aim to align both for better impact.")
return "\n- " + "\n- ".join(suggestions)
def generate_personacoach_report(emotions, transcript):
report = "## ๐Ÿ“ **Your PersonaCoach Report**\n---\n\n"
report += "### ๐Ÿ—’๏ธ **What You Said:**\n"
report += f"> _{transcript.strip()}_\n\n"
label_map = {
'hap': '๐Ÿ˜Š happiness', 'sad': '๐Ÿ˜” sadness', 'neu': '๐Ÿ˜ neutral',
'ang': '๐Ÿ˜  anger', 'fea': '๐Ÿ˜จ fear', 'dis': '๐Ÿคข disgust', 'sur': '๐Ÿ˜ฎ surprise'
}
for e in emotions:
e['emotion'] = label_map.get(e['label'], e['label'])
scores = [s['score'] for s in emotions]
top_score = max(scores)
conf_score = int(top_score * 100)
meaningful_emotions = [(e['emotion'], e['score']) for e in emotions if e['score'] >= 0.2]
emotion_labels = [e[0] for e in meaningful_emotions]
dominant_emotion = emotion_labels[0] if emotion_labels else "neutral"
report += f"### ๐ŸŽฏ **Tone Strength:**\n- Your tone scored **{conf_score}/100** in clarity.\n\n"
report += "### ๐Ÿ—ฃ๏ธ **Emotion & Delivery:**\n"
if meaningful_emotions:
emotions_str = ", ".join([f"**{label}** ({score:.2f})" for label, score in meaningful_emotions])
report += f"- Emotionally, your voice showed: {emotions_str}\n"
else:
report += "- Your tone wasnโ€™t clearly expressive. Try reading with a bit more emphasis or emotion.\n"
filler_words = ["um", "uh", "like", "you know", "so", "actually", "basically", "literally"]
words = transcript.lower().split()
total_words = len(words)
filler_count = sum(words.count(fw) for fw in filler_words)
filler_ratio = filler_count / total_words if total_words > 0 else 0
report += "\n### ๐Ÿ’ฌ **Pausing Style (e.g., 'um', 'like', 'you know'):**\n"
report += f"- You used **{filler_count}** hesitation phrases out of **{total_words}** words.\n"
if filler_ratio > 0.06:
report += "- Try pausing instead of using fillers โ€” it builds stronger presence.\n"
elif filler_ratio > 0.03:
report += "- A few slipped in. Practice holding space with silence instead.\n"
else:
report += "- Great fluency โ€” you stayed focused and controlled.\n"
report += "\n### โœ… **What You're Doing Well:**\n"
if top_score >= 0.75 and filler_ratio < 0.03:
report += "- Confident tone and smooth delivery โ€” keep it up!\n"
else:
report += "- Youโ€™re on track. Keep refining tone and pacing.\n"
report += "\n### ๐Ÿงญ **Next Moves:**\n"
report += generate_next_moves(dominant_emotion, conf_score, transcript) + "\n"
return report
def analyze_audio(audio_path):
result = whisper_model.transcribe(audio_path)
transcript = result['text']
emotion_results = emotion_classifier(audio_path)
labels = [r['label'] for r in emotion_results]
scores = [r['score'] for r in emotion_results]
fig = create_emotion_chart(labels, scores)
report = generate_personacoach_report(emotion_results, transcript)
return transcript, fig, report
with gr.Blocks(title="SPEAK: PersonaCoach", theme=gr.themes.Soft()) as app:
gr.Markdown("""
<div style="text-align:center; margin-bottom: 1rem;">
<h1 style="font-size: 2.2rem; margin-bottom: 0.2rem;">๐ŸŽค SPEAK: PersonaCoach</h1>
<p style="color: gray;">Your smart voice reflection tool โ€” assess tone, confidence, and delivery</p>
</div>
""", elem_id="header")
with gr.Row():
with gr.Column(scale=4):
audio_input = gr.Audio(type="filepath", label="๐ŸŽง Upload Your Voice (.wav)", elem_id="upload-audio")
with gr.Column(scale=1, min_width=120):
analyze_btn = gr.Button("๐Ÿ” Analyze", size="sm", elem_id="analyze-btn")
gr.Markdown("## ๐Ÿง  Results", elem_id="results-header")
with gr.Row(equal_height=True):
with gr.Column(scale=2):
feedback_output = gr.Markdown(label="๐Ÿ“„ PersonaCoach Feedback", elem_id="report-section")
with gr.Column(scale=1):
emotion_plot = gr.Plot(label="๐ŸŽญ Emotion Chart", elem_id="chart")
analyze_btn.click(
fn=analyze_audio,
inputs=audio_input,
outputs=[gr.Textbox(visible=False), emotion_plot, feedback_output]
)
app.launch()