Generate HTTP POST form with multipart-form-data without curl - http-headers

Question

So I'm trying to generate HTTP POST form in my embedded application. However I get server 400 error that indicates that something is wrong with my post. I do not have any curl-like libraries, or such, so I need to form the post header from scratch.
const static char *post_header = "POST /v1/avs/speechrecognizer/recognize HTTP/1.1\r\n\
Host: access-alexa-na.amazon.com\r\n\
Authorization: Bearer %s\r\n\
Content-Type: multipart/form-data; boundary=BOUNDARY1234\r\n\
Transfer-Encoding: chunked\r\n\
Content-Length: %d\r\n\
\r\n\r\n\
--BOUNDARY1234\r\n\
Content-Disposition: form-data; name=\"metadata\"\r\n\
Content-Type: application/json; charset=UTF-8\r\n\
\r\n\
{\"messageHeader\": {},\"messageBody\": {\"profile\": \"alexa-close-talk\",\"locale\": \"en-us\",\"format\": \"audio/L16; rate=16000; channels=1\"}}\r\n\
\r\n\r\n\
--BOUNDARY1234\r\n\
Content-Disposition: form-data; name=\"audio\"\r\n\
Content-Type: audio/L16; rate=16000; channels=1\r\n\n";
After the last "\n" I have the wav header and payload itself. I don't have null termination between the wav header and the last request header content. Altough I've tried it and it doesn't seem to make any difference.
My authentication token should be OK (I've verified it with curl). I've used these scripts (https://miguelmota.com/blog/alexa-voice-service-with-curl/) and Amazon documentation as a base. The blogpost has a script that generates multipart payload and it's identical (compared binary dumps) to mine. My only obvious questionmarks are the first part of the query:
"POST /v1/avs/speechrecognizer/recognize HTTP/1.1\r\n\
Host: access-alexa-na.amazon.com\r\n\
Authorization: Bearer %s\r\n\
Content-Type: multipart/form-data; boundary=BOUNDARY1234\r\n\
Transfer-Encoding: chunked\r\n\
Content-Length: %d\r\n\
\r\n\r\n\"
and the curl call with especially the --data-binary part. Should it effect the request body shomehow?
curl -X POST \
-H "Authorization: Bearer ${TOKEN}" \
-H "Content-Type: multipart/form-data; boundary=${BOUNDARY}" \
--data-binary #multipart_body.txt \
https://access-alexa-na.amazon.com/v1/avs/speechrecognizer/recognize \
> response.txt
Any ideas anyone? I'm gettin a bit frustrated with this.
EDIT 1: Just to clarify. The total size of the data is about 200kbytes with the audio data included. The header size with the token is about 1200bytes. I'm sending the stuff in 1k blobs and I get the error after 4k or so. So I don't manage to send the whole thing before the server responds with the error. Also some of the similar cases in Amazon side indicates that 400 in this case points to problem with the header. However they aren't manually forming the posts so I cannot see the whole thing anywhere.
EDIT2:
Also as this is chunked data, I wonder how it affects this?
I mean if I fe chunk the header into parts defined by the --BOUNDARY1234 and max of 512 bytes, how would that work? I mean fe:
--BOUNDARY1234\r\n\ Content-Disposition: form-data; name=\"metadata\"\r\n\ Content-Type: application/json;
charset=UTF-8\r\n\ \r\n\ {\"messageHeader\": {},\"messageBody\":
{\"profile\": \"alexa-close-talk\",\"locale\": \"en-us\",\"format\":
\"audio/L16; rate=16000; channels=1\"}}\r\n\ \r\n\r\n\
Should the there be chunk size right in the start of the transfer before the --BOUNDARY1234 or does the "Content-Disposition" or "Content-Type" affect this somehow? Or should I add the chunk size only to binary payload? Problem here is that the max send block with my HW is 1k. And the total header size is ~1,5k.