File size: 9,639 Bytes
d448add
 
db46bfb
 
 
 
 
 
 
 
d448add
db46bfb
 
 
d448add
db46bfb
d448add
db46bfb
d448add
 
 
db46bfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d448add
 
db46bfb
 
 
d448add
db46bfb
 
 
 
 
 
 
 
 
 
 
 
 
d448add
db46bfb
 
 
d448add
db46bfb
 
 
 
 
 
 
 
 
 
 
 
d448add
db46bfb
d448add
db46bfb
 
d448add
db46bfb
 
d448add
 
db46bfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d448add
db46bfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d448add
db46bfb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d448add
db46bfb
d448add
 
 
db46bfb
 
 
d448add
db46bfb
d448add
db46bfb
 
 
d448add
db46bfb
 
 
d448add
db46bfb
 
d448add
db46bfb
 
 
 
 
 
 
d448add
 
 
db46bfb
 
 
 
 
 
 
d448add
 
 
 
 
db46bfb
d448add
db46bfb
d448add
db46bfb
 
 
 
 
 
 
 
 
d448add
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import streamlit as st
import torch
import scipy.io.wavfile
from transformers import (
    AutoTokenizer, 
    AutoModelForCausalLM, 
    pipeline,
    AutoProcessor, 
    MusicgenForConditionalGeneration
)

# ---------------------------------------------------------------------
# Page Configuration
# ---------------------------------------------------------------------
st.set_page_config(
    page_icon="🎧",
    layout="wide",
    page_title="Radio Imaging Audio Generator - Llama & MusicGen",
    initial_sidebar_state="expanded",
)

# ---------------------------------------------------------------------
# Custom CSS for a Vibrant UI
# ---------------------------------------------------------------------
CUSTOM_CSS = """
<style>
body {
    background-color: #F8FBFE;
    color: #1F2937;
    font-family: 'Segoe UI', Tahoma, sans-serif;
}
h1, h2, h3, h4, h5, h6 {
    color: #3B82F6;
}
.stButton>button {
    background-color: #3B82F6 !important;
    color: #FFFFFF !important;
    border-radius: 8px !important;
    font-size: 16px !important;
}
.sidebar .sidebar-content {
    background: #E0F2FE;
}
.material-card {
    border: 1px solid #D1D5DB;
    border-radius: 8px;
    padding: 1rem;
    margin-bottom: 1rem;
    background-color: #ffffff;
}
.footer-note {
    text-align: center; 
    opacity: 0.6; 
    font-size: 14px; 
    margin-top: 30px;
}
</style>
"""
st.markdown(CUSTOM_CSS, unsafe_allow_html=True)

# ---------------------------------------------------------------------
# Header Section
# ---------------------------------------------------------------------
st.markdown(
    """
    <h1>Radio Imaging Audio Generator <span style="font-size: 24px; color: #F59E0B;">(Beta)</span></h1>
    <p style='font-size:18px;'>
        Generate custom radio imaging audio, ads, and promo tracks with Llama & MusicGen!
    </p>
    """,
    unsafe_allow_html=True
)
st.markdown("---")

# ---------------------------------------------------------------------
# Instructions Section in an Expander
# ---------------------------------------------------------------------
with st.expander("📘 How to Use This Web App"):
    st.markdown(
        """
        1. **Enter your prompt**: Describe the type of audio you need (e.g., an energetic 15-second jingle for a pop radio promo).
        2. **Generate Description**: Let Llama 2 (or another open-source model) refine your prompt into a creative script.
        3. **Generate Audio**: Pass that script to MusicGen to get a custom audio file.
        4. **Playback & Download**: Listen to your new track and download it for further editing.
        
        **Tips**:
        - Keep descriptions short & specific for best results.
        - If the Llama model is too large, switch to a smaller open-source model or try a GPU-based environment.
        - If you see errors about model permissions, ensure you’ve accepted the license on Hugging Face.
        """
    )

# ---------------------------------------------------------------------
# Sidebar: Model Selection & Options
# ---------------------------------------------------------------------
with st.sidebar:
    st.header("🔧 Model Config")
    # Llama 2 chat model from Hugging Face
    llama_model_id = st.text_input(
        "Llama 2 Model ID on Hugging Face",
        value="meta-llama/Llama-2-7b-chat-hf",
        help="For example: meta-llama/Llama-2-7b-chat-hf (requires license acceptance)."
    )
    device_option = st.selectbox(
        "Hardware Device",
        ["auto", "cpu"],
        help="If running locally with a GPU, choose 'auto'. If you only have a CPU, pick 'cpu'."
    )

# ---------------------------------------------------------------------
# Prompt Input
# ---------------------------------------------------------------------
st.markdown("## ✍🏻 Write Your Brief / Concept")
prompt = st.text_area(
    "Describe the radio imaging or jingle you want to create. Include style, mood, duration, etc.",
    placeholder="e.g. 'An energetic 15-second pop jingle for a morning radio show, upbeat and fun...'"
)

# ---------------------------------------------------------------------
# Text Generation with Llama
# ---------------------------------------------------------------------
@st.cache_resource
def load_llama_pipeline(model_id: str, device: str):
    """
    Load the Llama or other open-source model as a text-generation pipeline.
    The user must have accepted the license for certain models like Llama 2.
    """
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForCausalLM.from_pretrained(
        model_id,
        torch_dtype=torch.float16 if device == "auto" else torch.float32,
        device_map=device
    )
    gen_pipeline = pipeline(
        "text-generation",
        model=model,
        tokenizer=tokenizer,
        device_map=device
    )
    return gen_pipeline

def generate_description(user_prompt: str, pipeline_gen):
    """
    Use the pipeline to create a refined description for MusicGen.
    """
    # Instruction format for Llama 2 chat
    # or simpler prompt if it's not a chat model
    system_prompt = (
        "You are a helpful assistant specialized in creative advertising scripts and radio imaging. "
        "Refine the user's short concept into a more detailed, creative script. "
        "Keep it concise, but highlight any relevant tone, instruments, or style to guide music generation."
    )
    
    # We'll feed a combined prompt
    combined_prompt = f"{system_prompt}\nUser request: {user_prompt}\nYour refined script:"
    
    # Generate text
    result = pipeline_gen(
        combined_prompt,
        max_new_tokens=200,
        do_sample=True,
        temperature=0.7
    )
    # Extract generated text (some models output extra tokens or the entire prompt again)
    generated_text = result[0]["generated_text"]
    
    # Attempt to cut out the system prompt if it reappears
    # Just a heuristic: find the last occurrence of "script:" or any relevant marker
    if "script:" in generated_text.lower():
        generated_text = generated_text.split("script:")[-1].strip()
    
    # Optional: add a sign-off or credit line
    generated_text += "\n\n(Generated by Radio Imaging Audio Generator - Llama Edition)"
    return generated_text

# Button: Generate Description
if st.button("📄 Refine Description with Llama"):
    if not prompt.strip():
        st.error("Please provide a brief concept before generating a description.")
    else:
        with st.spinner("Generating a refined description..."):
            try:
                pipeline_llama = load_llama_pipeline(llama_model_id, device_option)
                refined_text = generate_description(prompt, pipeline_llama)
                st.session_state['refined_prompt'] = refined_text
                st.success("Description successfully refined!")
                st.write(refined_text)
                st.download_button(
                    "📥 Download Description",
                    refined_text,
                    file_name="refined_description.txt"
                )
            except Exception as e:
                st.error(f"Error while generating with Llama: {e}")

st.markdown("---")

# ---------------------------------------------------------------------
# MusicGen: Generate Audio
# ---------------------------------------------------------------------
@st.cache_resource
def load_musicgen_model():
    """Load and cache the MusicGen model and processor."""
    mg_model = MusicgenForConditionalGeneration.from_pretrained("facebook/musicgen-small")
    mg_processor = AutoProcessor.from_pretrained("facebook/musicgen-small")
    return mg_model, mg_processor

if st.button("▶ Generate Audio with MusicGen"):
    if 'refined_prompt' not in st.session_state or not st.session_state['refined_prompt']:
        st.error("Please generate or have a refined description first.")
    else:
        descriptive_text = st.session_state['refined_prompt']
        with st.spinner("Generating your audio... This can take a moment."):
            try:
                musicgen_model, processor = load_musicgen_model()
                # Use the refined prompt as input
                inputs = processor(
                    text=[descriptive_text],
                    padding=True,
                    return_tensors="pt"
                )
                audio_values = musicgen_model.generate(**inputs, max_new_tokens=512)
                sampling_rate = musicgen_model.config.audio_encoder.sampling_rate

                # Save & display the audio
                audio_filename = "radio_imaging_output.wav"
                scipy.io.wavfile.write(
                    audio_filename,
                    rate=sampling_rate,
                    data=audio_values[0, 0].numpy()
                )
                st.success("Audio successfully generated!")
                st.audio(audio_filename)
            except Exception as e:
                st.error(f"Error while generating audio: {e}")

# ---------------------------------------------------------------------
# Footer Section
# ---------------------------------------------------------------------
st.markdown("---")
st.markdown(
    "<div class='footer-note'>"
    "✅ Built with Llama 2 & MusicGen · "
    "Created for radio imaging producers · "
    "Feedback welcome at <a href='https://bilsimaging.com' target='_blank'>Bilsimaging</a>!"
    "</div>",
    unsafe_allow_html=True
)
# Hide Streamlit's default menu and footer if you wish
st.markdown("<style>#MainMenu {visibility: hidden;} footer {visibility: hidden;}</style>", unsafe_allow_html=True)