微软翻译语音URL文本长度限制突破方法咨询

Great question—let’s break down what’s happening here and how you can replicate the Bing Translate website’s long-text speech capability.

First, the Official Limit Context #

The direct https://www.bing.com/tspeak endpoint you’re using does have a built-in text length limit (usually 200-400 words, as you noticed). This is intentional: this endpoint is designed for short, on-the-fly speech requests (like translating a single sentence and hearing it immediately).

The Bing Translate website doesn’t use this endpoint directly for long text. Instead, it uses a more sophisticated backend workflow that splits your text into smaller, valid chunks, processes each one separately, and then stitches the audio fragments together seamlessly for playback.

How to Replicate the Website’s Long-Text Speech Generation #

To handle 5000-word texts like the website does, you’ll need to implement a similar chunk-and-stitch workflow. Here’s a step-by-step breakdown:

1. Split Your Long Text into Manageable Chunks #

You’ll want to split the text at natural semantic breaks to avoid awkward pauses or broken words in the speech output. Good split points include:

• Periods, exclamation marks, or question marks
• Paragraph breaks
• Line breaks

Keep each chunk under the endpoint’s limit (aim for 250-300 words max to stay safe). For example, in Python, you could use a simple function to split text by sentences:

def split_long_text(text, max_chunk_length=250):
    sentences = text.split('. ')
    chunks = []
    current_chunk = ""
    
    for sentence in sentences:
        if len(current_chunk) + len(sentence) <= max_chunk_length:
            current_chunk += sentence + '. '
        else:
            chunks.append(current_chunk.strip())
            current_chunk = sentence + '. '
    
    if current_chunk:
        chunks.append(current_chunk.strip())
    return chunks

2. Fetch Audio for Each Chunk #

For each chunk, make a request to the tspeak endpoint with the chunk’s text. Important notes here:

• Keep all other parameters (language, format, IG, IID) consistent across requests to ensure the same voice and audio quality.
• Be mindful of request frequency—don’t spam the endpoint too quickly, as this could trigger rate limiting.
• The IG parameter is a session identifier that may expire over time. If you start getting errors, grab a fresh IG value from the Bing Translate website’s network traffic (check the tspeak requests in your browser’s dev tools).

3. Merge the Audio Fragments #

Once you have all the audio files (MP3s), you’ll need to merge them into a single, continuous audio file. How you do this depends on your tech stack:

• Backend (Python): Use libraries like pydub to load each MP3, concatenate them, and export the result:

  from pydub import AudioSegment
  
  audio_chunks = []
  for chunk in text_chunks:
      # Fetch the MP3 content for the chunk and save to a temp file or load directly
      audio = AudioSegment.from_mp3("temp_chunk.mp3")
      audio_chunks.append(audio)
  
  merged_audio = sum(audio_chunks)
  merged_audio.export("full_text_speech.mp3", format="mp3")
  

• Frontend (JavaScript): Use the Web Audio API to load each audio blob, decode them, and concatenate the audio buffers into a single buffer for playback.

Alternative: Use Microsoft’s Official Speech Service #

If you want a more reliable, supported solution without handling chunking yourself, consider using the Azure Cognitive Services Speech SDK. This is the official, enterprise-grade service behind Bing Translate’s speech capabilities, and it supports long-text speech synthesis out of the box (with much higher limits). It’s a paid service, but there’s a free tier for testing.

内容的提问来源于stack exchange，提问作者Ave