Article abstract for podcast like firstory/poddtoppen/小宇宙FM

Code

#pak::pkg_install('tuneR')
library(ellmer)
library(tidyverse)
library(srt)
library(openxlsx)
library(readxl)
library(lares)
library(tuneR)
library(stringr)
library(rvest)
library(av)

Step 1: download mp3

get all mp3 link from episodes_link

Code

i="https://open.firstory.me/story/cm8mfryux2mj001socwiefcw4"
page_content <- read_html(i) 

html_list=html_elements(page_content, css = "script") |> 
    html_text() |> paste(collapse = " ")

Code

a=html_list |> str_extract_all( 'audioUrl(.+?).mp3')
input_link=a |> unlist()|> str_extract_all( 'https:(.+?).mp3') |> unlist()

Code

output='output.mp3'
download.file(url = input_link, destfile = output)

Code

list.files()

check mp3 duration

Code

# Load the MP3 file
mp3_file <- readMP3("output.mp3" )

# Get the duration in seconds
duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60
duration_mins

trim mp3 if needed

Code

trim_mp3(
  'output.mp3',
  start_time = 1,
  end_time = 600,
  overwrite = FALSE,
  ext = "mp3",
  quiet = FALSE
)

# # output file:output_trim.mp3

Code

# Load the MP3 file
mp3_file <- readMP3("output_trimmed.mp3" ) 
# Get the duration in seconds
(length(mp3_file@left) / mp3_file@samp.rate)/60

get mp3 link from episodes_link

Code

i="https://poddtoppen.se/podcast/1670019206/%E6%A5%AD%E5%8B%99%E7%94%A8%E5%A8%81%E5%A3%AB%E5%BF%8C%E6%8C%87%E5%8D%97/ep69%E8%AA%B0%E4%B8%8D%E6%84%9B%E9%A4%BE%E9%85%92"
page_content <- read_html(i)

Code

html_list=html_elements(page_content, css = "div.episodes__list__item") |> 
    html_attr("data-audio")

html_list

Code

#a=html_list |> str_extract_all( 'data-audio=(.+?).mp3')
#input_link=a |> unlist()|> str_extract_all( 'https:(.+?).mp3') |> unlist()
input_link=html_list

Code

output='output.mp3'
download.file(url = input_link, destfile = output)

Code

list.files()

check mp3 duration

Code

# Load the MP3 file
mp3_file <- readMP3("output.mp3" )

# Get the duration in seconds
duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60
duration_mins

trim mp3 if needed

Code

# trim_mp3(
#   'output.mp3',
#   start_time = 1,
#   end_time = 600,
#   overwrite = FALSE,
#   ext = "mp3",
#   quiet = FALSE
# )

# # output file:output_trim.mp3

Code

# Load the MP3 file
mp3_file <- readMP3("output_trimmed.mp3" ) 
# Get the duration in seconds
(length(mp3_file@left) / mp3_file@samp.rate)/60

get mp3 link from episodes_link

https://www.xiaohongshu.com/user/profile/64034a70000000001001d5e7?xsec_token=ABhaT1fRn3eiAgLpQycyyc-rKPx7wiVd43VWTkC9cznbM=&xsec_source=pc_search

https://www.xiaohongshu.com/user/profile/5cd7cac4000000001000da74?xsec_token=ABZS9k_q_Yve0bqEqs8EOUgYOp0yLuRdlXTecaQlHcTt4=&xsec_source=pc_search

https://www.xiaohongshu.com/user/profile/63bfc800000000002502cbdc?xsec_token=ABY8GKfUekMVr5qho4f8G9Dr6MgOMgQ5B6H-lOvb_usAA=&xsec_source=pc_note

Code

i='https://www.xiaoyuzhoufm.com/podcast/670e1e6477e599f60e29dacf'
page <- read_html(i)

Code

# 定位所有class="jsx-7bbe0f84186f1998"的li元素
li_elements <- page %>%
  html_nodes("li.jsx-7bbe0f84186f1998")

# 提取每个li元素中的class="jsx-744662fb2f5b91b6 card"的a元素的href值
href_links <- li_elements %>%
  html_nodes("a.jsx-744662fb2f5b91b6.card") %>%
  html_attr("href")

# 添加前缀，构成完整的音频文件URL
full_links <- paste0("https://www.xiaoyuzhoufm.com", href_links)

Code

# download the lastest one
new_url=full_links[1]
new_url
# 访问新的URL并解析
audio_page <- read_html(new_url)
    
# 提取音频标题，property="og:title"的meta元素的content值
title <- audio_page %>%
html_nodes('meta[property="og:title"]') %>%
html_attr("content")
    
# 处理标题中的特殊字符
safe_title <- str_replace_all(title, "[\\/:,*?\"<>|]", "_")
safe_title <- str_replace_all(safe_title, "[\\s]", "_")
    
# 提取音频下载链接，property="og:audio"的meta元素的content值
audio_url <- audio_page %>%
html_nodes('meta[property="og:audio"]') %>%
html_attr("content")

Code

safe_title
audio_url

Code

# Convert m4a to WAV
av_audio_convert("your_audio.m4a", "your_audio.wav")

Code

#output_file_name=paste0(safe_title ,'output.mp3')
output_file_name='output.m4a'
download.file(url = audio_url, destfile = output_file_name)

convert m4a to mp3

Code

av_audio_convert('output.m4a', 'output.mp3')

Code

list.files()

check mp3 duration

Code

# Load the MP3 file
mp3_file <- readMP3('output.mp3')

# Get the duration in seconds
duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60
duration_mins

trim mp3 if needed

Code

# trim_mp3(
#   'output.mp3',
#   start_time = 1,
#   end_time = 600,
#   overwrite = FALSE,
#   ext = "mp3",
#   quiet = FALSE
# )

# # output file:output_trim.mp3

Code

# # Load the MP3 file
# mp3_file <- readMP3("output_trimmed.mp3" ) 
# # Get the duration in seconds
# (length(mp3_file@left) / mp3_file@samp.rate)/60

Step 2: using mlx_whisper model to get text from audio(mp3,m4a)

run audio_txt.py

import mlx_whisper
from whisper.utils import get_writer
speech_file="output.mp3"
# Using mlx-community/whisper-large-v3-turbo model
result = mlx_whisper.transcribe(speech_file, 
                                path_or_hf_repo="mlx-community/whisper-large-v3-turbo",
                                word_timestamps=True
                                )

srt_writer = get_writer("srt",'.')
srt_writer(result,'text.srt')

srt_writer = get_writer("txt",'.')
srt_writer(result,'text.txt')

Code

import os 
os.system("python3.11 audio_txt.py")

output is text.txt and text.srt

Step 3: Using gemini model to summary

model gemini-2.5-pro-exp-03-25 or gemini-2.0-flash

define model

Code

chat_gemini_model<- chat_gemini(
  system_prompt = "你是一个中文，英文，威士忌专家",
  turns = NULL,
  # base_url = "https://generativelanguage.googleapis.com/v1beta",
  api_key = keyring::key_get("google_ai_api_key"),
  model = "gemini-2.0-flash",
  #api_args = list(),
  #echo = NULL
)
chat_gemini_model

Code

#testing model connection
chat_result=chat_gemini_model$chat("hello")
chat_result

Run model

Code

srt_txt0=read_srt('text.srt')
srt_txt2=srt_txt0$subtitle|> as.character()

Code

length(srt_txt2)

Code

prompt_text=paste0('请给以下文字作500字内摘要：',srt_txt2)
summary_text=chat_gemini_model$chat(prompt_text)

Code

summary_text |> tibble() |> write_delim('summary.txt')

Step 4: correct summary

define model

Code

chat_gemini_model<- chat_gemini(
  system_prompt = "你是一个中文和英文的威士忌专家",
  turns = NULL,
  # base_url = "https://generativelanguage.googleapis.com/v1beta",
  api_key = keyring::key_get("google_ai_api_key"),
  model = "gemini-2.0-flash",
  #api_args = list(),
  #echo = NULL
)
chat_gemini_model

Run model

Code

prompt_text=paste0('请更正以下文字的错别字，并且改正胡云为壶云，希游记为嬉游忌,Wish Jokey为WhisJockey,不要空白行',summary_text)
correct_summary_text=chat_gemini_model$chat(prompt_text)

Code

correct_summary_text |> str_replace_all('\n\n','\n')|> tibble() |> write_delim('correct_summary2.txt')

Step 5: summary str text

define model

Code

chat_gemini_model<- chat_gemini(
  system_prompt = "你是一个中文和英文的威士忌专家",
  turns = NULL,
  # base_url = "https://generativelanguage.googleapis.com/v1beta",
  api_key = keyring::key_get("google_ai_api_key"),
  model = "gemini-2.0-flash",
  #model = "gemini-2.5-pro-exp-03-25",
  #api_args = list(),
  #echo = NULL
)
chat_gemini_model

Run model

Code

srt_txt_format=read.delim('text.srt')

Code

prompt_text=paste0('下面的内容是srt文档。请按每5分钟做一个摘要，再更正以下文字，胡云为壶云，希游记为嬉游忌,Wish Jokey为WhisJockey',srt_txt_format)
correct_summary_text=chat_gemini_model$chat(prompt_text)

Code

correct_summary_text|> tibble() |> write_delim('correct_srt_summary.txt')

--- title: "使用AI给播客语音转文字并作摘要" subtitle: "Using AI to create Summary for podcast" author: "Tony D" date: "2025-03-28" categories: - AI - R - Python execute: warning: false error: false eval: false image: 'images/my screenshots.png' --- Article abstract for podcast like firstory/poddtoppen/小宇宙FM using mlx_whisper to Transcribe and gemini 2.0 flash to summarize ```{r} #pak::pkg_install('tuneR') library(ellmer) library(tidyverse) library(srt) library(openxlsx) library(readxl) library(lares) library(tuneR) library(stringr) library(rvest) library(av) ``` # Step 1: download mp3 ::: panel-tabset ## from firstory ### get all mp3 link from episodes_link ```{r} i="https://open.firstory.me/story/cm8mfryux2mj001socwiefcw4" page_content <- read_html(i) html_list=html_elements(page_content, css = "script") |> html_text() |> paste(collapse = " ") ``` ```{r} a=html_list |> str_extract_all( 'audioUrl(.+?).mp3') input_link=a |> unlist()|> str_extract_all( 'https:(.+?).mp3') |> unlist() ``` ```{r} output='output.mp3' download.file(url = input_link, destfile = output) ``` ```{r} list.files() ``` ### check mp3 duration ```{r} # Load the MP3 file mp3_file <- readMP3("output.mp3" ) # Get the duration in seconds duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60 duration_mins ``` ### trim mp3 if needed ```{r} trim_mp3( 'output.mp3', start_time = 1, end_time = 600, overwrite = FALSE, ext = "mp3", quiet = FALSE ) # # output file:output_trim.mp3 ``` ```{r} # Load the MP3 file mp3_file <- readMP3("output_trimmed.mp3" ) # Get the duration in seconds (length(mp3_file@left) / mp3_file@samp.rate)/60 ``` ## from poddtoppen ### get mp3 link from episodes_link ```{r} i="https://poddtoppen.se/podcast/1670019206/%E6%A5%AD%E5%8B%99%E7%94%A8%E5%A8%81%E5%A3%AB%E5%BF%8C%E6%8C%87%E5%8D%97/ep69%E8%AA%B0%E4%B8%8D%E6%84%9B%E9%A4%BE%E9%85%92" page_content <- read_html(i) ``` ```{r} html_list=html_elements(page_content, css = "div.episodes__list__item") |> html_attr("data-audio") html_list ``` ```{r} #a=html_list |> str_extract_all( 'data-audio=(.+?).mp3') #input_link=a |> unlist()|> str_extract_all( 'https:(.+?).mp3') |> unlist() input_link=html_list ``` ```{r} output='output.mp3' download.file(url = input_link, destfile = output) ``` ```{r} list.files() ``` ### check mp3 duration ```{r} # Load the MP3 file mp3_file <- readMP3("output.mp3" ) # Get the duration in seconds duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60 duration_mins ``` ### trim mp3 if needed ```{r} # trim_mp3( # 'output.mp3', # start_time = 1, # end_time = 600, # overwrite = FALSE, # ext = "mp3", # quiet = FALSE # ) # # output file:output_trim.mp3 ``` ```{r} # Load the MP3 file mp3_file <- readMP3("output_trimmed.mp3" ) # Get the duration in seconds (length(mp3_file@left) / mp3_file@samp.rate)/60 ``` ## from xiaoyuzhou 小宇宙FM ### get mp3 link from episodes_link https://www.xiaohongshu.com/user/profile/64034a70000000001001d5e7?xsec_token=ABhaT1fRn3eiAgLpQycyyc-rKPx7wiVd43VWTkC9cznbM=&xsec_source=pc_search https://www.xiaohongshu.com/user/profile/5cd7cac4000000001000da74?xsec_token=ABZS9k_q_Yve0bqEqs8EOUgYOp0yLuRdlXTecaQlHcTt4=&xsec_source=pc_search https://www.xiaohongshu.com/user/profile/63bfc800000000002502cbdc?xsec_token=ABY8GKfUekMVr5qho4f8G9Dr6MgOMgQ5B6H-lOvb_usAA=&xsec_source=pc_note ```{r} i='https://www.xiaoyuzhoufm.com/podcast/670e1e6477e599f60e29dacf' page <- read_html(i) ``` ```{r} # 定位所有class="jsx-7bbe0f84186f1998"的li元素 li_elements <- page %>% html_nodes("li.jsx-7bbe0f84186f1998") # 提取每个li元素中的class="jsx-744662fb2f5b91b6 card"的a元素的href值 href_links <- li_elements %>% html_nodes("a.jsx-744662fb2f5b91b6.card") %>% html_attr("href") # 添加前缀，构成完整的音频文件URL full_links <- paste0("https://www.xiaoyuzhoufm.com", href_links) ``` ```{r} # download the lastest one new_url=full_links[1] new_url # 访问新的URL并解析 audio_page <- read_html(new_url) # 提取音频标题，property="og:title"的meta元素的content值 title <- audio_page %>% html_nodes('meta[property="og:title"]') %>% html_attr("content") # 处理标题中的特殊字符 safe_title <- str_replace_all(title, "[\\/:,*?\"<>|]", "_") safe_title <- str_replace_all(safe_title, "[\\s]", "_") # 提取音频下载链接，property="og:audio"的meta元素的content值 audio_url <- audio_page %>% html_nodes('meta[property="og:audio"]') %>% html_attr("content") ``` ```{r} safe_title audio_url ``` ```{r} # Convert m4a to WAV av_audio_convert("your_audio.m4a", "your_audio.wav") ``` ```{r} #output_file_name=paste0(safe_title ,'output.mp3') output_file_name='output.m4a' download.file(url = audio_url, destfile = output_file_name) ``` ### convert m4a to mp3 ```{r} av_audio_convert('output.m4a', 'output.mp3') ``` ```{r} list.files() ``` ### check mp3 duration ```{r} # Load the MP3 file mp3_file <- readMP3('output.mp3') # Get the duration in seconds duration_mins <- (length(mp3_file@left) / mp3_file@samp.rate)/60 duration_mins ``` ### trim mp3 if needed ```{r} # trim_mp3( # 'output.mp3', # start_time = 1, # end_time = 600, # overwrite = FALSE, # ext = "mp3", # quiet = FALSE # ) # # output file:output_trim.mp3 ``` ```{r} # # Load the MP3 file # mp3_file <- readMP3("output_trimmed.mp3" ) # # Get the duration in seconds # (length(mp3_file@left) / mp3_file@samp.rate)/60 ``` ::: # Step 2: using mlx_whisper model to get text from audio(mp3,m4a) run audio_txt.py ```python {{< include audio_txt.py >}} ``` ```{python} import os os.system("python3.11 audio_txt.py") ``` output is text.txt and text.srt # Step 3: Using gemini model to summary model gemini-2.5-pro-exp-03-25 or gemini-2.0-flash ## define model ```{r} chat_gemini_model<- chat_gemini( system_prompt = "你是一个中文，英文，威士忌专家", turns = NULL, # base_url = "https://generativelanguage.googleapis.com/v1beta", api_key = keyring::key_get("google_ai_api_key"), model = "gemini-2.0-flash", #api_args = list(), #echo = NULL ) chat_gemini_model ``` ```{r} #testing model connection chat_result=chat_gemini_model$chat("hello") chat_result ``` ```{r} chat_gemini_model$get_turns(include_system_prompt = TRUE) ``` ## Run model ```{r} srt_txt0=read_srt('text.srt') srt_txt2=srt_txt0$subtitle|> as.character() ``` ```{r} length(srt_txt2) ``` ```{r} prompt_text=paste0('请给以下文字作500字内摘要：',srt_txt2) summary_text=chat_gemini_model$chat(prompt_text) ``` ```{r} summary_text |> tibble() |> write_delim('summary.txt') ``` # Step 4: correct summary ## define model ```{r} chat_gemini_model<- chat_gemini( system_prompt = "你是一个中文和英文的威士忌专家", turns = NULL, # base_url = "https://generativelanguage.googleapis.com/v1beta", api_key = keyring::key_get("google_ai_api_key"), model = "gemini-2.0-flash", #api_args = list(), #echo = NULL ) chat_gemini_model ``` ## Run model ```{r} prompt_text=paste0('请更正以下文字的错别字，并且改正胡云为壶云，希游记为嬉游忌,Wish Jokey为WhisJockey,不要空白行',summary_text) correct_summary_text=chat_gemini_model$chat(prompt_text) ``` ```{r} correct_summary_text |> str_replace_all('\n\n','\n')|> tibble() |> write_delim('correct_summary2.txt') ``` # Step 5: summary str text ## define model ```{r} chat_gemini_model<- chat_gemini( system_prompt = "你是一个中文和英文的威士忌专家", turns = NULL, # base_url = "https://generativelanguage.googleapis.com/v1beta", api_key = keyring::key_get("google_ai_api_key"), model = "gemini-2.0-flash", #model = "gemini-2.5-pro-exp-03-25", #api_args = list(), #echo = NULL ) chat_gemini_model ``` ## Run model ```{r} srt_txt_format=read.delim('text.srt') ``` ```{r} prompt_text=paste0('下面的内容是srt文档。请按每5分钟做一个摘要，再更正以下文字，胡云为壶云，希游记为嬉游忌,Wish Jokey为WhisJockey',srt_txt_format) correct_summary_text=chat_gemini_model$chat(prompt_text) ``` ```{r} correct_summary_text|> tibble() |> write_delim('correct_srt_summary.txt') ```