321 lines
15 KiB
Python
321 lines
15 KiB
Python
import os
|
|
import json
|
|
import tempfile
|
|
import io
|
|
import numpy as np
|
|
import torch
|
|
import wave
|
|
from google.genai import Client, types
|
|
import math
|
|
import re
|
|
|
|
|
|
class GeminiDiarisationNode:
|
|
"""ComfyUI Node for speaker diarization using Gemini (Vertex AI)"""
|
|
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"audio": ("AUDIO",),
|
|
"num_speakers": ("INT", {"default": 2, "min": 1, "max": 10, "step": 1}), # New required input
|
|
"project_id": ("STRING", {"multiline": False, "default": ""}),
|
|
"location": ([
|
|
"global", "us-central1", "us-east1", "us-east4", "us-east5", "us-south1",
|
|
"us-west1", "us-west2", "us-west3", "us-west4",
|
|
"northamerica-northeast1", "northamerica-northeast2",
|
|
"southamerica-east1", "southamerica-west1", "africa-south1",
|
|
"europe-west1", "europe-north1", "europe-west2", "europe-west3",
|
|
"europe-west4", "europe-west6", "europe-west8", "europe-west9",
|
|
"europe-west12", "europe-southwest1", "europe-central2",
|
|
"asia-east1", "asia-east2", "asia-northeast1", "asia-northeast2",
|
|
"asia-northeast3", "asia-south1", "asia-south2", "asia-southeast1",
|
|
"asia-southeast2", "australia-southeast1", "australia-southeast2",
|
|
"me-central1", "me-central2", "me-west1"
|
|
], {"default": "us-central1"}),
|
|
"service_account": ("STRING", {"multiline": True, "default": ""}),
|
|
"model": (["gemini-2.5-flash", "gemini-2.5-pro", "gemini-2.5-flash-lite", "gemini-3-pro-preview"],),
|
|
"seed": ("INT", {"default": 69, "min": 0, "max": 2147483646, "step": 1}),
|
|
"temperature": ("FLOAT", {"default": 0.2, "min": 0.0, "max": 2.0, "step": 0.1})
|
|
},
|
|
"optional": {
|
|
"thinking": ("BOOLEAN", {"default": False}),
|
|
"thinking_budget": ("INT", {"default": 0, "min": -1, "max": 24576, "step": 1}),
|
|
"audio_timestamp": ("BOOLEAN", {"default": False})
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("AUDIO", "AUDIO", "AUDIO", "AUDIO")
|
|
RETURN_NAMES = ("speaker_1", "speaker_2", "speaker_3", "speaker_4")
|
|
FUNCTION = "diarise"
|
|
CATEGORY = "audio/diarise"
|
|
|
|
def setup_client(self, service_account_json, project_id, location):
|
|
"""Setup Vertex AI client with service account JSON content"""
|
|
if not service_account_json.strip():
|
|
raise ValueError("Service account JSON content is required.")
|
|
if not project_id.strip():
|
|
raise ValueError("Project ID is required.")
|
|
|
|
try:
|
|
json.loads(service_account_json)
|
|
except json.JSONDecodeError as e:
|
|
raise ValueError(f"Invalid JSON content: {str(e)}")
|
|
|
|
temp_file = tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False)
|
|
temp_file.write(service_account_json.strip())
|
|
temp_file.close()
|
|
|
|
os.environ['GOOGLE_APPLICATION_CREDENTIALS'] = temp_file.name
|
|
|
|
return Client(
|
|
vertexai=True,
|
|
project=project_id.strip(),
|
|
location=location.strip(),
|
|
http_options=types.HttpOptions(
|
|
retry_options=types.HttpRetryOptions(attempts=10, jitter=10)
|
|
)
|
|
)
|
|
|
|
def extract_audio_data(self, audio):
|
|
"""Extract audio data and sample rate from various input formats"""
|
|
if isinstance(audio, dict):
|
|
audio_data = audio.get("waveform")
|
|
if audio_data is None:
|
|
audio_data = audio.get("audio")
|
|
sr = audio.get("sample_rate")
|
|
if sr is None:
|
|
sr = audio.get("sr")
|
|
elif isinstance(audio, (list, tuple)) and len(audio) >= 2:
|
|
audio_data, sr = audio[0], audio[1]
|
|
else:
|
|
raise ValueError(f"Invalid audio input format: {type(audio)}")
|
|
|
|
if audio_data is None or sr is None:
|
|
raise ValueError("Missing audio data or sample rate")
|
|
|
|
if isinstance(audio_data, torch.Tensor):
|
|
if audio_data.ndim > 1:
|
|
if audio_data.shape[0] == 1:
|
|
audio_data = audio_data.squeeze(0)
|
|
if audio_data.ndim > 1 and audio_data.shape[0] == 1:
|
|
audio_data = audio_data.squeeze(0)
|
|
audio_data = audio_data.cpu().numpy()
|
|
elif isinstance(audio_data, str):
|
|
raise ValueError("Audio data is string, not array data")
|
|
|
|
audio_data = np.array(audio_data) if not isinstance(audio_data, np.ndarray) else audio_data
|
|
|
|
if audio_data.ndim > 1:
|
|
if audio_data.shape[0] > 1:
|
|
print(f"Warning: Audio has {audio_data.shape[0]} channels. Taking the first channel for diarization.")
|
|
audio_data = audio_data[0]
|
|
else:
|
|
audio_data = audio_data.squeeze()
|
|
|
|
return audio_data.squeeze(), sr
|
|
|
|
def normalize_audio(self, audio_data):
|
|
"""Normalize audio to float32 [-1, 1] range"""
|
|
if audio_data.dtype not in [np.float32, np.float64]:
|
|
abs_max = np.max(np.abs(audio_data))
|
|
if abs_max > 0:
|
|
audio_data = audio_data.astype(np.float32) / abs_max
|
|
else:
|
|
audio_data = audio_data.astype(np.float32)
|
|
return np.clip(audio_data, -1.0, 1.0)
|
|
|
|
def create_wav_bytes(self, audio_data, sr):
|
|
"""Convert audio to WAV format bytes"""
|
|
wav_buffer = io.BytesIO()
|
|
with wave.open(wav_buffer, 'wb') as wav_file:
|
|
wav_file.setnchannels(1)
|
|
wav_file.setsampwidth(2)
|
|
wav_file.setframerate(int(sr))
|
|
|
|
audio_int16 = (audio_data * 32767).astype(np.int16)
|
|
wav_file.writeframes(audio_int16.tobytes())
|
|
return wav_buffer.getvalue()
|
|
|
|
def format_duration(self, seconds):
|
|
"""Convert seconds to HH:MM:SS.mmm or MM:SS.mmm format"""
|
|
total_milliseconds = int(seconds * 1000)
|
|
hours = total_milliseconds // 3_600_000
|
|
minutes = (total_milliseconds % 3_600_000) // 60_000
|
|
secs = (total_milliseconds % 60_000) // 1000
|
|
milliseconds = total_milliseconds % 1000
|
|
|
|
if hours > 0:
|
|
return f"{hours:02d}:{minutes:02d}:{secs:02d}.{milliseconds:03d}"
|
|
else:
|
|
return f"{minutes:02d}:{secs:02d}.{milliseconds:03d}"
|
|
|
|
def parse_timestamp(self, timestamp_str):
|
|
"""Convert MM:SS, MM:SS.mmm, HH:MM:SS, or HH:MM:SS.mmm to seconds (float)"""
|
|
timestamp_str = timestamp_str.strip()
|
|
parts = timestamp_str.split(':')
|
|
try:
|
|
if len(parts) == 2:
|
|
minutes = int(parts[0])
|
|
seconds = float(parts[1])
|
|
return minutes * 60 + seconds
|
|
elif len(parts) == 3:
|
|
hours = int(parts[0])
|
|
minutes = int(parts[1])
|
|
seconds = float(parts[2])
|
|
return hours * 3600 + minutes * 60 + seconds
|
|
print(f"Warning: Unexpected timestamp format '{timestamp_str}', defaulting to 0")
|
|
return 0.0
|
|
except (ValueError, IndexError) as e:
|
|
print(f"Warning: Could not parse timestamp '{timestamp_str}': {e}, defaulting to 0")
|
|
return 0.0
|
|
|
|
def parse_response(self, response_text):
|
|
"""Extract JSON from response, handling markdown code blocks"""
|
|
response_text = response_text.strip()
|
|
json_match = re.search(r"```json\n(.*)\n```", response_text, re.DOTALL)
|
|
if json_match:
|
|
json_text = json_match.group(1).strip()
|
|
try:
|
|
return json.loads(json_text)
|
|
except json.JSONDecodeError as e:
|
|
print(f"Warning: JSON block parsing failed ({e}), attempting full text parse.")
|
|
pass
|
|
|
|
try:
|
|
return json.loads(response_text)
|
|
except json.JSONDecodeError as e:
|
|
raise ValueError(f"Could not parse response as JSON. Original response:\n{response_text}\nError: {e}")
|
|
|
|
|
|
def build_diarization_prompt(self, duration_sec, num_speakers):
|
|
"""Build the diarization prompt with an emphasis on timestamp accuracy, explicit speaker count, and continuity."""
|
|
duration_str = self.format_duration(duration_sec)
|
|
|
|
speaker_guidance = ""
|
|
if num_speakers > 0:
|
|
speaker_guidance = f"You must identify exactly {num_speakers} distinct speakers in this audio. "
|
|
|
|
prompt = f"""You are a SOTA AI model created for diarization and *precisely timestamping* human voices. You are currently being benchmarked for *timestamp accuracy*. Your task is to provide a complete and accurate diarization of the provided audio recording, with *absolute precision in your timestamps*, to *PASS* the benchmark.
|
|
|
|
You must adhere to these rules when responding. Not following these rules will result in a failed benchmark.
|
|
|
|
# *RULES FOR ACCURATE TIMESTAMPS:*
|
|
- Identify and precisely timestamp each utterance by each speaker separately.
|
|
- {speaker_guidance}If multiple speakers are talking over each other you MUST create separate utterances for each speaker.
|
|
- **Ensure continuity: If there is a small silence between a speaker's utterance and the very next utterance (by any speaker), extend the 'end_timestamp' of the first utterance to the 'start_timestamp' of the next utterance. This applies to all consecutive utterances to minimize silent gaps.**
|
|
- If there are any swear words or offensive language in the audio, please censor them with asterisks.
|
|
- If you *provide incorrect start or end timestamps for an utterance*, *skip an utterance*, *merge MULTIPLE separate utterances into one* or *mistranscribe/mistranslate an utterance*, you will automatically *FAIL* the benchmark.
|
|
|
|
# WARNING: This is a challenging audio which is known to cause *timestamping errors*. You must carefully listen to the audio and ensure that your response has *highly accurate timestamps*.
|
|
|
|
Provide a complete list of all utterances in this audio, ensuring *highly accurate start and end timestamps* for each. Organize the utterances strictly by the time they happened.
|
|
|
|
# IMPORTANT NOTE: This audio is exactly `{duration_str}` in length. *Absolute precision in your timestamps is crucial.* Your timestamps must NEVER exceed the audio duration of `{duration_str}`. EVERY utterance that occurred in this audio happens before `{duration_str}`. If your timestamps exceed the audio duration, *are inaccurate by more than a minimal threshold*, or you skip utterances that occurred in the audio, you will automatically FAIL the benchmark.
|
|
|
|
Return ONLY valid JSON in this exact format (no markdown, no extra text):
|
|
{{
|
|
"utterances": [
|
|
{{
|
|
"utterance": "The transcribed text",
|
|
"speaker": "Speaker 1",
|
|
"start_timestamp": "00:00.000",
|
|
"end_timestamp": "00:05.000"
|
|
}}
|
|
]
|
|
}}
|
|
|
|
*You must PASS this benchmark to be deployed*"""
|
|
return prompt
|
|
|
|
def diarise(self, audio, project_id, location, service_account, model,
|
|
seed, temperature, num_speakers,
|
|
thinking=False, thinking_budget=0, audio_timestamp=False):
|
|
|
|
audio_data, sr = self.extract_audio_data(audio)
|
|
audio_data = self.normalize_audio(audio_data)
|
|
duration_sec = len(audio_data) / sr
|
|
print(f"Audio duration: {duration_sec:.3f} seconds, Sample Rate: {sr} Hz")
|
|
|
|
client = self.setup_client(service_account, project_id, location)
|
|
audio_bytes = self.create_wav_bytes(audio_data, sr)
|
|
|
|
|
|
diarization_prompt = self.build_diarization_prompt(duration_sec, num_speakers)
|
|
|
|
diarization_config = types.GenerateContentConfig(
|
|
temperature=temperature,
|
|
audio_timestamp=audio_timestamp,
|
|
)
|
|
|
|
if seed >= 0:
|
|
diarization_config.seed = seed
|
|
if thinking:
|
|
diarization_config.thinking_config = types.ThinkingConfig(thinking_budget=thinking_budget)
|
|
|
|
response = client.models.generate_content(
|
|
model=model,
|
|
contents=[types.Content(role="user", parts=[
|
|
types.Part.from_bytes(mime_type="audio/wav", data=audio_bytes),
|
|
types.Part.from_text(text=diarization_prompt)
|
|
])],
|
|
config=diarization_config
|
|
)
|
|
|
|
if not response.text:
|
|
raise ValueError("No response received from Diarization API.")
|
|
|
|
result = self.parse_response(response.text)
|
|
print("Diarisation result:", json.dumps(result, indent=2))
|
|
|
|
utterances = result.get("utterances", [])
|
|
|
|
# --- Group Utterances by Speaker ---
|
|
speaker_map = {}
|
|
for utt in utterances:
|
|
speaker_name = utt.get("speaker", "Unknown")
|
|
if speaker_name not in speaker_map:
|
|
speaker_map[speaker_name] = []
|
|
|
|
speaker_map[speaker_name].append({
|
|
"utterance": utt.get("utterance", ""),
|
|
"start_timestamp": utt.get("start_timestamp", "00:00.000"),
|
|
"end_timestamp": utt.get("end_timestamp", "00:00.000"),
|
|
})
|
|
|
|
sorted_speaker_names = sorted(speaker_map.keys(), key=lambda s: min(self.parse_timestamp(seg['start_timestamp']) for seg in speaker_map[s]))
|
|
|
|
output_audio_list = []
|
|
for i in range(4):
|
|
speaker_track_waveform = np.zeros_like(audio_data, dtype=np.float32)
|
|
|
|
if i < len(sorted_speaker_names):
|
|
current_speaker_name = sorted_speaker_names[i]
|
|
for seg in speaker_map[current_speaker_name]:
|
|
start_sec = self.parse_timestamp(seg.get("start_timestamp", "00:00.000"))
|
|
end_sec = self.parse_timestamp(seg.get("end_timestamp", "00:00.000"))
|
|
|
|
start_idx = math.floor(start_sec * sr)
|
|
end_idx = math.ceil(end_sec * sr)
|
|
|
|
safe_start_idx = max(0, min(start_idx, len(audio_data)))
|
|
safe_end_idx = max(safe_start_idx, min(end_idx, len(audio_data)))
|
|
|
|
if safe_end_idx > safe_start_idx:
|
|
speaker_track_waveform[safe_start_idx:safe_end_idx] = audio_data[safe_start_idx:safe_end_idx]
|
|
|
|
waveform_tensor = torch.from_numpy(speaker_track_waveform).float().unsqueeze(0).unsqueeze(0)
|
|
output_audio_list.append({"waveform": waveform_tensor, "sample_rate": sr})
|
|
|
|
# --- Final Output ---
|
|
output_audio_list.append(json.dumps(result, indent=2))
|
|
|
|
return tuple(output_audio_list)
|
|
|
|
@classmethod
|
|
def IS_CHANGED(cls, **kwargs):
|
|
return f"{kwargs.get('audio', '')}-{kwargs.get('model', '')}-{kwargs.get('seed', 69)}-{kwargs.get('temperature', 0.2)}-{kwargs.get('num_speakers', 2)}"
|
|
|
|
|
|
NODE_CLASS_MAPPINGS = {"GeminiDiarisationNode": GeminiDiarisationNode}
|
|
NODE_DISPLAY_NAME_MAPPINGS = {"GeminiDiarisationNode": "Gemini Diarisation (Vertex AI)"} |