video-narrator（视频解说员）

SenseAudio Video Narrator

Create professional narration audio for videos with timing-aware segmentation, natural delivery, and editor-friendly exports.

What This Skill Does

• Generate narration audio synchronized to script timestamps
• Match narration style to video genre such as documentary or tutorial
• Control pacing with official TTS parameters and text break markers
• Create multiple narration takes with different voices or styles
• Export audio segments and merged narration tracks for editing workflows

Credential and Dependency Rules

• Read the API key from SENSEAUDIO_API_KEY.
• Send auth only as Authorization: Bearer <API_KEY>.
• Do not place API keys in query parameters, logs, or saved examples.
• If Python helpers are used, this skill expects python3, requests, and pydub.
• pydub is used only for optional local audio assembly and mixing.

Official TTS Constraints

Use the official SenseAudio TTS rules summarized below:

• HTTP endpoint: POST https://api.senseaudio.cn/v1/t2a_v2
• Model: SenseAudio-TTS-1.0
• Max text length per request: 10000 characters
• voice_setting.voice_id is required
• voice_setting.speed range: 0.5-2.0
• voice_setting.pitch range: -12 to 12
• Optional audio formats: mp3, wav, pcm, flac
• Optional sample rates: 8000, 16000, 22050, 24000, 32000, 44100
• Optional MP3 bitrates: 32000, 64000, 128000, 256000
• Optional channels: 1 or 2
• extra_info.audio_length returns segment duration in milliseconds
• Inline break markup such as <break time=500> is supported in text

Recommended Workflow

1. Prepare the script:

• Split narration into timestamped segments.
• Keep each segment comfortably below the 10000 character limit.

1. Choose a voice and pacing profile:

• Pick a voice_id and tune speed, pitch, and optional vol.
• Use shorter segments when timing precision matters.

1. Generate audio segments:

• Call the TTS API for each segment.
• Decode data.audio from hex before saving.
• Capture extra_info.audio_length for timeline metadata.

1. Assemble the narration track locally:

• Use pydub to position clips on a silent master track.
• Keep per-segment files for easier editor import and retiming.

1. Validate timing against the video:

• Leave small gaps when natural pacing is needed.
• Adjust segment boundaries instead of overusing extreme speed values.

Minimal Timed Narration Helper

import binascii
import os
import re

import requests

API_KEY = os.environ["SENSEAUDIO_API_KEY"]
API_URL = "https://api.senseaudio.cn/v1/t2a_v2"


def parse_timed_script(script):
    pattern = r"\[(\d{2}):(\d{2}):(\d{2})\]\s*(.+?)(?=\
\[|\Z)"
    segments = []
    for match in re.finditer(pattern, script, re.DOTALL):
        hours, minutes, seconds, text = match.groups()
        timestamp_ms = (int(hours) * 3600 + int(minutes) * 60 + int(seconds)) * 1000
        segments.append({"timestamp": timestamp_ms, "text": text.strip()})
    return segments


def synthesize_segment(text, voice_id, speed=1.0, pitch=0, vol=1.0):
    response = requests.post(
        API_URL,
        headers={
            "Authorization": f"Bearer {API_KEY}",
            "Content-Type": "application/json",
        },
        json={
            "model": "SenseAudio-TTS-1.0",
            "text": text,
            "stream": False,
            "voice_setting": {
                "voice_id": voice_id,
                "speed": speed,
                "pitch": pitch,
                "vol": vol,
            },
            "audio_setting": {
                "format": "mp3",
                "sample_rate": 32000,
                "bitrate": 128000,
                "channel": 2,
            },
        },
        timeout=60,
    )
    response.raise_for_status()
    data = response.json()
    return {
        "audio_bytes": binascii.unhexlify(data["data"]["audio"]),
        "duration_ms": data["extra_info"]["audio_length"],
        "trace_id": data.get("trace_id"),
    }

Local Assembly Pattern

from pydub import AudioSegment


def create_synced_narration(audio_segments, video_duration_ms):
    narration_track = AudioSegment.silent(duration=video_duration_ms)
    for segment in audio_segments:
        clip = AudioSegment.from_file(segment["file"])
        narration_track = narration_track.overlay(clip, position=segment["timestamp"])
    return narration_track

Style Presets

• Documentary: slower speed such as 0.95, neutral pitch
• Tutorial: speed near 1.0, slightly warmer pitch
• Commercial: modestly faster speed, slightly higher pitch

Prefer conservative tuning and script editing over extreme voice parameter changes.

Output Options

• Per-segment narration clips in mp3 or wav
• Timing metadata in json
• Merged narration track for video editors
• Optional alternate takes with different styles

Safety Notes

• Do not hardcode credentials.
• Do not assume local media tooling exists beyond what is declared here.
• Treat returned trace_id and generated narration assets as potentially sensitive production data.