Segment A Phone Call using LLM Gateway

1
import requests
2
import time
3
4
base_url = "https://api.assemblyai.com"
5
6
headers = {
7
    "authorization": "<YOUR_API_KEY>"
8
}
9
10
# Step 1: Transcribe and get VTT subtitles
11
with open("./my-audio.mp3", "rb") as f:
12
    response = requests.post(base_url + "/v2/upload", headers=headers, data=f)
13
14
upload_url = response.json()["upload_url"]
15
data = {"audio_url": upload_url}
16
17
response = requests.post(base_url + "/v2/transcript", json=data, headers=headers)
18
transcript_id = response.json()['id']
19
polling_endpoint = base_url + "/v2/transcript/" + transcript_id
20
21
while True:
22
    transcription_result = requests.get(polling_endpoint, headers=headers).json()
23
    if transcription_result['status'] == 'completed':
24
        break
25
    elif transcription_result['status'] == 'error':
26
        raise RuntimeError(f"Transcription failed: {transcription_result['error']}")
27
    else:
28
        time.sleep(3)
29
30
# Get VTT subtitles
31
vtt_response = requests.get(f"{polling_endpoint}/vtt", headers=headers)
32
vtt_content = vtt_response.text
33
34
# Step 2: Define phases and analyze with LLM Gateway
35
phases = ["Introduction", "Complaint", "Resolution", "Goodbye"]
36
37
prompt = f'''
38
Analyze the following transcript of a phone call conversation and divide it into the following phases:
39
{', '.join(phases)}
40
41
You will be given the transcript in the format of VTT captions.
42
43
For each phase:
44
1. Identify the start and end timestamps (in seconds)
45
2. Provide a brief summary of what happened in that phase
46
47
Format your response as a JSON object with the following structure:
48
{{
49
    "phases": [
50
        {{
51
            "name": "Phase Name",
52
            "start_time": start_time_in_seconds,
53
            "end_time": end_time_in_seconds,
54
            "summary": "Brief summary of the phase"
55
        }},
56
        ...
57
    ]
58
}}
59
60
Ensure that all parts of the conversation are covered by a phase, using "Other" for any parts that don't fit into the specified phases.
61
'''
62
63
llm_gateway_data = {
64
    "model": "claude-sonnet-4-5-20250929",
65
    "messages": [
66
        {"role": "user", "content": f"{prompt}\n\nVTT Transcript:\n{vtt_content}"}
67
    ],
68
    "max_tokens": 2000
69
}
70
71
response = requests.post(
72
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
73
    headers=headers,
74
    json=llm_gateway_data
75
)
76
77
result = response.json()["choices"][0]["message"]["content"]
78
print(result)

1
import requests
2
import time
3
4
base_url = "https://api.assemblyai.com"
5
6
headers = {
7
    "authorization": "<YOUR_API_KEY>"
8
}
9
10
with open("./my-audio.mp3", "rb") as f:
11
    response = requests.post(base_url + "/v2/upload",
12
                            headers=headers,
13
                            data=f)
14
15
upload_url = response.json()["upload_url"]
16
17
data = {
18
    "audio_url": upload_url # You can also use a URL to an audio or video file on the web
19
}
20
21
url = base_url + "/v2/transcript"
22
response = requests.post(url, json=data, headers=headers)
23
24
transcript_id = response.json()['id']
25
polling_endpoint = base_url + "/v2/transcript/" + transcript_id
26
27
while True:
28
    transcription_result = requests.get(polling_endpoint, headers=headers).json()
29
30
    if transcription_result['status'] == 'completed':
31
        print(f"Transcript ID:", transcript_id)
32
        break
33
34
    elif transcription_result['status'] == 'error':
35
        raise RuntimeError(f"Transcription failed: {transcription_result['error']}")
36
37
    else:
38
        time.sleep(3)
39
40
# Get VTT subtitles for timestamp information
41
vtt_response = requests.get(f"{polling_endpoint}/vtt", headers=headers)
42
vtt_content = vtt_response.text
43
44
with open(f"transcript_{transcript_id}.vtt", "w") as vtt_file:
45
    vtt_file.write(vtt_content)

1
phases = [
2
    "Introduction",
3
    "Complaint",
4
    "Resolution",
5
    "Goodbye"
6
]

1
prompt = f'''
2
Analyze the following transcript of a phone call conversation and divide it into the following phases:
3
{', '.join(phases)}
4
5
You will be given the transcript in the format of VTT captions.
6
7
For each phase:
8
1. Identify the start and end timestamps (in seconds)
9
2. Provide a brief summary of what happened in that phase
10
11
Format your response as a JSON object with the following structure:
12
{{
13
    "phases": [
14
        {{
15
            "name": "Phase Name",
16
            "start_time": start_time_in_seconds,
17
            "end_time": end_time_in_seconds,
18
            "summary": "Brief summary of the phase"
19
        }},
20
        ...
21
    ]
22
}}
23
24
Ensure that all parts of the conversation are covered by a phase, using "Other" for any parts that don't fit into the specified phases.
25
'''
26
27
llm_gateway_data = {
28
    "model": "claude-sonnet-4-5-20250929",
29
    "messages": [
30
        {"role": "user", "content": f"{prompt}\n\nVTT Transcript:\n{vtt_content}"}
31
    ],
32
    "max_tokens": 2000
33
}
34
35
response = requests.post(
36
    "https://llm-gateway.assemblyai.com/v1/chat/completions",
37
    headers=headers,
38
    json=llm_gateway_data
39
)
40
41
result = response.json()["choices"][0]["message"]["content"]
42
print(result)