Get value from cache if it exists and is not expired

Save value to cache with timestamp

Execute Google Custom Search with retry logic

Update progress display without clearing previous output

Stream version that yields tokens

* Create a test audio buffer with spoken content * This creates a simple sine wave that simulates audio content

Synchronous wrapper for generate_text

Try generating text using HTTP API

Render a slide template with context

* Create a sample audio buffer for testing * This creates a simple sine wave audio buffer

Clean and extract main content (static method)

Display research plan for debugging

Generate a file path for the cache key

Get completion using DeepSeek-R1

记录序列检测结果

Save current context to file

Analyze paper structure using VLM layout understanding

* Collect streaming response data

* Convert raw audio buffer to WAV format * @param {Buffer} audioBuffer - Raw PCM audio data * @param {Object} options - Audio format options

Create presentation narrative flow

Extract figures with captions using VLM assistance

Extract title, authors, affiliations using hybrid approach

Extract structured text with self-correction mechanism

Create config from environment variables

Generate Slidev markdown from structured content

Streaming version that yields tokens

* Get current provider information * @returns {Object} Current provider information

获取完整的注意力矩阵 - 包含所有生成token的注意力权重

* Reset VAD state

* Reset Silero VAD state tensors

* Transcribe audio using the configured ASR provider * @param {Buffer} audioBuffer - Raw audio data * @param {Object} options - Transcription op

更新注意力缓存

Update research status

Analyze gathered sources with parallel processing

More nuanced speed calculation

Actual VLM API call with retries

Preserve pacing markers while removing control codes

Remove header/footer and fix line breaks

Enforce slide count limits

Create high-contrast version with annotations

Create prompt for synthesizing multiple analyses