Audio

1
import OpenAI from 'openai';
2
import fs from "fs/promises";
3
4
const openai = new OpenAI({
5
  baseURL: 'https://api.staging.tram.ai.vn/v1',
6
  apiKey: '<TRAM_AI_API_KEY>',
7
});
8
9
async function encodeAudioToBase64(audioPath: string): Promise<string> {
10
  const audioBuffer = await fs.readFile(audioPath);
11
  return audioBuffer.toString("base64");
12
}
13
14
// Read and encode the audio file
15
const audioPath = "path/to/your/audio.wav";
16
const base64Audio = await encodeAudioToBase64(audioPath);
17
18
const result = await openai.chat.completions.create({
19
  model: "google/gemini-2.5-flash",
20
  messages: [
21
    {
22
      role: "user",
23
      content: [
24
        {
25
          type: "text",
26
          text: "Please transcribe this audio file.",
27
        },
28
        {
29
          type: "input_audio",
30
          inputAudio: {
31
            data: base64Audio,
32
            format: "wav",
33
          },
34
        },
35
      ],
36
    },
37
  ],
38
  stream: false,
39
});
40
41
console.log(result);

1
import requests
2
import json
3
import base64
4
5
url = "https://api.staging.tram.ai.vn/v1/chat/completions"
6
headers = {
7
    "Authorization": f"Bearer {API_KEY_REF}",
8
    "Content-Type": "application/json"
9
}
10
11
payload = {
12
    "model": "openai/gpt-4o-audio-preview",
13
    "messages": [
14
        {
15
            "role": "user",
16
            "content": "Say hello in a friendly tone."
17
        }
18
    ],
19
    "modalities": ["text", "audio"],
20
    "audio": {
21
        "voice": "alloy",
22
        "format": "wav"
23
    },
24
    "stream": True
25
}
26
27
# Audio output requires streaming — the response is delivered as SSE chunks
28
response = requests.post(url, headers=headers, json=payload, stream=True)
29
30
audio_data_chunks = []
31
transcript_chunks = []
32
33
for line in response.iter_lines():
34
    if not line:
35
        continue
36
    decoded = line.decode("utf-8")
37
    if not decoded.startswith("data: "):
38
        continue
39
    data = decoded[len("data: "):]
40
    if data.strip() == "[DONE]":
41
        break
42
    chunk = json.loads(data)
43
    delta = chunk["choices"][0].get("delta", {})
44
    audio = delta.get("audio", {})
45
    if audio.get("data"):
46
        audio_data_chunks.append(audio["data"])
47
    if audio.get("transcript"):
48
        transcript_chunks.append(audio["transcript"])
49
50
transcript = "".join(transcript_chunks)
51
print(f"Transcript: {transcript}")
52
53
# Combine and decode the base64 audio chunks, then save
54
full_audio_b64 = "".join(audio_data_chunks)
55
audio_bytes = base64.b64decode(full_audio_b64)
56
with open("output.wav", "wb") as f:
57
    f.write(audio_bytes)