ganga4364/STT_training_with_metadata

STT Training Data

This dataset is prepared for Speech-to-Text (STT) training. It contains audio files from different departments, each annotated with their transcription. The dataset is split by department and includes the duration of audio data per department.

Dataset Overview

• Total Audio Files: 1,391,020
• Total Audio Length: 1,576.76 hours
• Total Audio collected data: 18 OCT, 2024

Department-wise Statistics

Department	Number of Audio Files	Total Audio Length (hours)
STT_AB	174,525	224.92
STT_CS	150,310	219.51
STT_HS	87,088	96.27
STT_MV	112,756	193.27
STT_NS	241,007	318.28
STT_NW	97,145	137.82
STT_PC	73,894	78.90
STT_TT	454,295	307.78

Total Audio Length by Department

The total audio length of each department in hours:

• STT_AB: 224.92 hours
• STT_CS: 219.51 hours
• STT_HS: 96.27 hours
• STT_MV: 193.27 hours
• STT_NS: 318.28 hours
• STT_NW: 137.82 hours
• STT_PC: 78.90 hours
• STT_TT: 307.78 hours

Total Audio Length

The combined audio length across all departments is 1,576.76 hours.

Transcription Quality Grades:

Grade	Description
1	Transcribed only
2	Reviewed by the team lead
3	Reviewed by the quality control team

Departments and Metadata

Here is the structured table format for the metadata descriptions:

STT_Training Metadata

Column	Description
file_name	Structured name of the audio file, including department, project ID, segment ID, and time range.
uni	Tibetan transcript in Unicode.
wylie	Wylie transliteration of the Tibetan text.
url	URL link to the corresponding audio file.
dept	Department identifier (e.g., STT_NS).
grade	Quality grade of the transcription.
char_len	Character length of the transcript.
audio_len	Duration of the audio file in seconds.

STT_NS (Natural Speech) Metadata

Column	Description
Original ID	Unique identifier of the source, if available.
Speaker ID	Identifier for the speaker; multiple speakers are listed as a group.
Mic/Phone	Type of recording device used (e.g., microphone, phone).
Location	Geographical location of the speaker.
Topic	Subject matter of the audio.
A Short Scenario	Brief summary of the recording.
Speaker Name	Name of the speaker.
Email ID	Email address of the speaker.
Gender	Gender of the speaker.
Home Address	Speaker's home address.
Contact No	Contact number of the speaker.
Age	Age of the speaker.
Course	Associated course, if relevant.
Year	Year of the recording.
Nationality	Nationality of the speaker.
Birth Place	Speaker's birthplace.
Date of Birth	Date of birth of the speaker.
Native Language	First language of the speaker.
College	Associated institution of the speaker.

STT_AB (Audio Book) Metadata

Column	Description
Original ID	Unique identifier of the source, if available.
Collection	Collection or set to which the audio belongs.
Book Title	Title of the recorded book or text.
Category	Classification of the recording (e.g., text).
Recorder	Identifier of the person who recorded the audio.
Music	Indicates if background music is present (True/False).
Name	Speaker's name.
Gender	Gender of the speaker.
Age	Age of the speaker.
Place	Recording location.
Livelihood	Speaker's profession or occupation.
Birth Place	Birthplace of the speaker.
Social/Marital Status	Speaker’s social or marital status.
Education Level	Educational background of the speaker.

STT_NW (News) Metadata

Column	Description
Name of News	Title of the news segment.
Publishing Year	Year the news was published.
Media Organization	Media outlet responsible (e.g., TibetTV, VOA Tibetan).
Audio Quality	Quality rating of the audio.
Single Speaker	Indicates if a single speaker is present (True/False).
Country	Country of the news production.
Original ID	Unique identifier of the original source, if available.

STT_HS (Historical Speech) Metadata

Column	Description
file_name	Structured name of the audio file, including department, project ID, segment ID, and time range.
uni	Tibetan transcript in Unicode format, preserving the original speech content.
wylie	Wylie transliteration of the Tibetan transcript, providing a phonetic representation.
url	Direct URL link to the corresponding audio file stored on a cloud server.
dept	Department identifier, specifically for Historical Speech recordings (STT_HS).
grade	Quality grade of the transcription, indicating the level of review and quality assurance.
char_len	The character length of the transcript, representing the size of the text content.
audio_len	Duration of the audio file in seconds, indicating the length of the speech segment.
Publishing Year	Year the speech was recorded or published, providing historical context to the content.
Name	Name of the speaker, often with an alias to protect privacy.
Gender	Gender of the speaker, typically categorized as Male or Female.
Age Group	Age range of the speaker at the time of recording, categorized into groups.
Place of Origin	Region or cultural area the speaker originates from, such as Kham or Utsang.
Exiled Year	Year when the speaker went into exile, providing context to their personal history.
Original ID	Unique identifier for the source or original collection of the recording.

STT_CS Children Speech Metadata

Column	Description
Microphone_Type	Type of microphone used for recording, if available.
Recording_Type	The type of recording (e.g., classroom, outdoor, etc.).
Speakers ID	Unique identifier for the child speaker.
Birth Year	Year of birth of the child speaker, providing demographic context.
Gender	Gender of the child speaker, typically categorized as Male or Female.
CITY	City of residence of the child speaker, indicating their geographical background.
TOWN	Town of residence, if applicable, providing more specific location details.
school	Identifier for the school attended by the child, if available.
school name	Name of the school or educational institution attended by the child.
Birth Place	Birthplace of the child speaker, indicating their origin.
spoken tongue	Primary language spoken by the child, providing context for their linguistic background.
Original ID	Unique identifier for the source or original collection of the recording.
CLASS	Academic class or grade level of the child at the time of the recording.

STT_MV Movie Speech Metadata

Column	Description
Name of the movie	Title of the movie from which the dialogue is extracted.
Subtitle	Subtitle information for the dialogue, if available (e.g., "Chinese").
Original ID	Unique identifier for the source or original collection of the recording.

STT_PC Podcast Speech Metadata

Column	Description
Name of podcast	Title of the podcast episode, highlighting the main topic discussed.
Publishing year	Year the podcast episode was published, providing context to the content.
Channel	Name of the channel or platform where the podcast is broadcasted (e.g., "Tibetan Libr").
Audio Quality	Rating of the audio quality, indicating clarity and production standards (e.g., "Good").
Country	Country of origin or production for the podcast episode (e.g., "India").
Gender	Gender of the speaker in the podcast episode (e.g., "Male").
Age Group	Age range of the speaker at the time of recording (e.g., "60-70").
Original ID	Unique identifier for the source or original collection of the recording.

STT_TT Tibetan Teaching Talk Metadata

Column	Description
Teacher	Name of the teacher delivering the talk, often with an alias to protect privacy.
Topic	Subject matter discussed in the talk, providing context to the content.
books	Associated books or materials referenced in the talk, if applicable.
Episode	Episode number or identifier for the talk series.
Original ID	Unique identifier for the source or original collection of the recording.

---